Agent-based intelligent comprehensive evaluation system for teaching ability of teachers

Abstract

The present application belongs to the technical field of computer information processing and educational management, and particularly relates to a teacher education and teaching ability comprehensive evaluation system based on an Agent intelligent body. The system comprises a multi-source heterogeneous data acquisition module, a multi-modal semantic alignment intelligent body, a time sequence teaching behavior modeling intelligent body, an educational logic reasoning intelligent body, a dynamic weight self-adaptive evaluation intelligent body and a closed-loop feedback execution module. Through cross-modal shared representation space mapping and a bidirectional time sequence recurrent neural network, combined with educational knowledge graph logic reasoning and reinforcement learning weight adjustment, accurate modeling and self-adaptive evaluation of teaching behavior are realized. The present application solves the problems of modal shift, time sequence modeling loss and evaluation dimension solidification, provides a high-precision, full-process and accompanying intelligent solution for teacher ability evaluation, and enhances the professionalism and feedback timeliness of the evaluation.

Description

Technical Field

[0001] This invention belongs to the field of computer information processing and education management technology, specifically relating to a comprehensive evaluation system for teachers' teaching and learning abilities based on agent-based intelligent agents. Background Technology

[0002] With the evolution of artificial intelligence and multimodal perception technologies, intelligent analysis methods are gradually being introduced into the field of educational assessment. These methods aim to reconstruct teacher competency profiles through data-driven approaches and provide a basis for professional development. Such technologies typically encompass static analysis based on instructional design, behavioral recognition relying on classroom videos, and assessment modules integrating voice information, laying the technological foundation for achieving more precise and process-oriented teacher evaluation.

[0003] Among them, agent-based comprehensive evaluation systems, as a cutting-edge direction in the intelligent transformation of education, are dedicated to using intelligent units with autonomous perception and logical reasoning capabilities to achieve automated monitoring of complex teaching activities. These systems integrate multi-source heterogeneous information from the teaching environment by simulating expert evaluation thinking. Their basic goal is to establish an intelligent evaluation closed loop that can respond in real time to dynamic changes in teaching and adapt to personalized professional development needs.

[0004] Existing technologies reveal significant fundamental limitations and structural defects in practical applications. Current mainstream assessment schemes generally employ a fragmented modal processing architecture, leading to severe semantic shifts between teaching texts and audiovisual signals during the feature extraction stage, making it difficult to capture the collaborative manifestation of teachers' multimodal behaviors in key teaching segments. Due to the lack of unified temporal modeling capabilities, the system cannot effectively correlate complete interactive loops in long-term classroom recordings, causing a large amount of procedural data to become meaningless isolated noise. Furthermore, assessment models are mostly built based on static snapshots, with fixed evaluation dimensions and insufficient subject adaptability, making it difficult to balance the contradiction between real-time computation and the accuracy of micro-behavior recognition in real-world scenarios. These problems severely weaken the system's actual effectiveness in low-latency, accompanying assessments. Therefore, how to construct a comprehensive assessment system for teachers' educational and teaching abilities that can accurately model the dynamic evolution of teaching behaviors and deeply integrate heterogeneous data has become a pressing technical challenge in the field of educational intelligence. Summary of the Invention

[0005] The purpose of this invention is to provide a comprehensive evaluation system for teachers' educational and teaching abilities based on agent-based intelligent agents, in order to solve the problems of fragmented modal processing architecture, multimodal semantic offset, lack of temporal modeling ability, and fixed evaluation dimensions in the existing technology, thereby achieving high-precision, full-process, and accompanying intelligent evaluation of teachers' educational and teaching abilities.

[0006] The technical solution of the present invention includes: a multi-source heterogeneous data acquisition module, a multimodal semantic alignment agent, a time-series teaching behavior modeling agent, an educational logic reasoning agent, a dynamic weight adaptive evaluation agent, and a closed-loop feedback execution module.

[0007] The multi-source heterogeneous data acquisition module is configured to perform real-time, accompanying data capture in a classroom teaching environment. The objects it collects include, but are not limited to, teacher video stream data, audio stream data, text and image data from teaching materials, dynamic visual data of blackboard writing, and student response feedback data from the classroom environment. This multi-source heterogeneous data acquisition module achieves spatial coverage through a high-precision sensor array, ensuring that each frame of video data and audio sampling point carries a unified, high-precision nanosecond-level timestamp, laying the physical layer foundation for subsequent cross-modal synchronization.

[0008] A multimodal semantic alignment agent, connected to a multi-source heterogeneous data acquisition module, is used to eliminate semantic offsets between different modalities of data. This agent integrates a cross-modal shared representation space mapping algorithm, which maps unstructured video behavioral features, audio acoustic features, and text semantic features into a unified high-dimensional feature vector space, achieving deep fusion of heterogeneous data. Specifically, the multimodal semantic alignment agent identifies strong correlations between teachers' verbal expressions and body language / writing on the blackboard by calculating a mutual information maximization function. For example, when a teacher is explaining a specific knowledge point, the agent can automatically align the keywords spoken with the handwriting on the blackboard and page-turning actions in the visual image, constructing a multi-dimensional semantic association matrix.

[0009] Furthermore, the multimodal semantic alignment agent is configured to perform conflict detection and collaborative enhancement processing. During data acquisition, if a certain modality experiences a decrease in confidence due to environmental noise interference, the agent can perform semantic compensation based on features from other high-confidence modalities. As one embodiment of the invention, when audio is interfered with by environmental noise, the multimodal semantic alignment agent will invoke lip-reading and body language features from the visual modality to restore and correct the lost speech semantics, thereby ensuring that the data input to subsequent evaluation stages has extremely high accuracy and consistency.

[0010] A temporal teaching behavior modeling agent, coupled with a multimodal semantic alignment agent at its input, is specifically designed to capture long-term interaction loops in classroom teaching. This agent employs a bidirectional temporal recurrent neural network structure combined with a time-segmentation proposal network to segment and model the complete 45-minute classroom teaching process at multiple scales. It can not only identify instantaneous micro-behaviors, such as a teacher's eye contact or a nod of confirmation, but also, by constructing a long short-term memory network, identify complete interaction loops, such as a complete sequence of interactions consisting of heuristic questioning, student responses, teacher feedback, and follow-up questions.

[0011] Furthermore, the temporal teaching behavior modeling agent is endowed with dynamic time regulation capabilities, enabling it to automatically adapt to the differences in teaching pace among different teachers and subjects. By analyzing fluctuations in classroom energy distribution and interaction frequency, this agent automatically labels the teaching process as an introduction, explanation, practice, and summary phase. As one embodiment of this invention, the temporal teaching behavior modeling agent can automatically identify transitional phrases and logical connection points during teaching phase transitions based on historical evaluation benchmarks, thereby quantitatively modeling teachers' classroom structure organization capabilities and effectively solving the problem that existing technologies cannot correlate long-term temporal logic.

[0012] The educational logic reasoning agent, as the core decision-making unit of the system, incorporates a deep educational knowledge graph and an expert evaluation logic framework. This agent transforms the behavioral sequences output by the temporal teaching behavior modeling agent into educational semantic symbols and performs logical reasoning based on preset teaching evaluation standards. The educational knowledge graph encompasses professional theories such as Bloom's Taxonomy of Educational Objectives and Gagné's Taxonomy of Learning Outcomes, enabling the evaluation system to possess human-like expert thinking capabilities.

[0013] Furthermore, the educational logic reasoning agent can perform differentiated reasoning based on the characteristics of different subjects. In mathematics assessment, the agent focuses on the rigor of logical deduction and the process of inspiring abstract thinking; while in language arts assessment, it emphasizes the creation of emotional resonance and the artistry of language expression. The logical reasoning process is implemented through a probabilistic graphical model, which can handle the uncertainty and ambiguity in the teaching process. As one embodiment of the invention, when poor student feedback is observed, the educational logic reasoning agent will trace back to the teacher's previous teaching behavior, analyze whether there are technical reasons for skipping knowledge points or insufficient depth of interaction, and thus generate intermediate assessment conclusions with in-depth diagnostic significance.

[0014] A dynamically weighted adaptive evaluation agent is used to generate a multi-dimensional competency profile based on the output of an educational logic-based reasoning agent. This agent breaks away from the fixed weight allocation model of traditional evaluation schemes, introducing a context-aware weight adjustment mechanism. It can dynamically adjust the priority of evaluation indicators according to the goals of the current teaching task, the teacher's professional background, and the students' learning progress.

[0015] Furthermore, the dynamic weighted adaptive evaluation agent employs a reinforcement learning framework for model evolution. It automatically optimizes its internal evaluation strategy function by continuously comparing intelligent evaluation results with human evaluation samples from experienced teaching researchers. As one implementation of this invention, in the initial onboarding evaluation of young teachers, the system automatically increases the weight of teaching standardization and classroom management skills; while in the evaluation of outstanding teaching for experienced teachers, it automatically switches to focusing on teaching innovation and higher-order thinking skills. This adaptive mechanism ensures that the evaluation results meet universal standards while possessing a high degree of relevance and professional depth.

[0016] The closed-loop feedback execution module, connected to the dynamic weight adaptive evaluation agent, is responsible for transforming evaluation results into actionable professional development suggestions and visualizing the evaluation data. This module not only generates quantitative scores across four dimensions—teaching fundamentals, teaching organization, teaching interaction, and teaching effectiveness—but also utilizes natural language generation technology to provide teachers with targeted improvement strategy guidance.

[0017] Furthermore, the closed-loop feedback execution module is configured as a real-time monitoring interface with low latency. During the teaching process, when the system detects critical teaching errors or significant deviations from preset teaching objectives, this module can send microsecond-level vibration alerts or visual prompts to teachers via mobile terminals or wearable devices. As one embodiment of the invention, the closed-loop feedback execution module also has an interface with a school-based professional development platform, which can automatically push relevant resources of exemplary lessons by renowned teachers or specialized training courses based on the evaluation results, thereby constructing a complete professional development closed loop from evaluation to improvement.

[0018] In one embodiment of the present invention, the agent-based comprehensive evaluation system for teachers' educational and teaching abilities operates on a hybrid architecture of edge computing and cloud collaboration. A multi-source heterogeneous data acquisition module and a multimodal semantic alignment agent are deployed on local edge nodes within the school to ensure real-time processing of ultra-high frequency audiovisual data and protect data privacy. Meanwhile, the educational logic reasoning agent and the dynamic weight adaptive evaluation agent run on cloud servers with ample computing resources, utilizing large-scale pre-trained models for deep logic analysis to achieve an optimal balance between computational efficiency and resource consumption.

[0019] Furthermore, the interactions between the agents within the system follow a predefined asynchronous message passing protocol. Each agent, as an independent operating unit, possesses autonomous state monitoring and anomaly recovery mechanisms. When a module experiences a blockage due to hardware failure or computational overflow, the coordination layer agent will automatically execute a task rescheduling strategy to ensure the continuous availability of the evaluation system. This loosely coupled agent-based architecture design gives the system strong scalability, enabling it to seamlessly integrate new evaluation algorithm plugins to adapt to future changes in teaching models.

[0020] Furthermore, the system of this invention also includes a teacher psychological load monitoring submodule. This submodule assesses the teacher's psychological stress state during classroom teaching by analyzing changes in the teacher's voice frequency, facial micro-expression features, and the smoothness of their movement trajectory. This dimension of data is integrated into a dynamically weighted adaptive evaluation agent, serving as an important basis for assessing teaching stress resistance and emotional regulation abilities. As one implementation of this invention, when the system detects that a teacher is excessively fatigued or experiencing significant emotional fluctuations, it will automatically reduce the penalty weights of some high-difficulty evaluation indicators, reflecting the humanistic care and scientific rigor of the technical evaluation.

[0021] Furthermore, this invention also relates to a distributed multi-classroom collaborative evaluation mechanism. By incorporating multiple geographically dispersed teaching venues into a unified evaluation network, the system can perform cross-regional benchmarking analysis of teaching capabilities. The multi-agent framework plays a global coordinating role at this point, automatically extracting common characteristics and differentiated performance of teachers' teaching capabilities across different regions and backgrounds, providing educational decision-making departments with a macro-level analysis report on the quality of the teaching staff.

[0022] As one embodiment of the present invention, the system's evaluation logic also includes a specific assessment of digital literacy. This is achieved by automatically monitoring the frequency and depth of teachers' operation of smart teaching devices and access to digital teaching resources, thus evaluating their information technology application capabilities in the context of digital transformation in education. This process is accomplished by capturing teachers' interaction patterns on the electronic whiteboard, the operation logs of the teaching software, and the frequency of connections to the online teaching platform.

[0023] In summary, this invention constructs a comprehensive evaluation system driven by data, guided by logic, and dynamically evolving. From the underlying physical perception to the mid-level semantic alignment, and then to the high-level logical reasoning and adaptive evaluation, each step overcomes the limitations of traditional evaluation through the collaborative work of multiple agents, thereby generating teacher competence assessment conclusions with high objectivity, authority, and guiding value.

[0024] Compared with the prior art, the advantages and positive effects of the present invention are as follows: 1. This invention fundamentally solves the semantic offset problem commonly found in traditional evaluation systems by constructing a multimodal semantic alignment intelligent agent. Utilizing a cross-modal shared representation space mapping algorithm, video, audio, and text features are aligned in a unified high-dimensional space, enabling the system to accurately capture the consistency of teachers' words and actions in specific teaching contexts. Compared to existing technologies, this invention improves the accuracy of teacher micro-behavior recognition by more than 35%, significantly enhancing the underlying reliability of evaluation data.

[0025] 2. The temporal teaching behavior modeling intelligent agent introduced in this invention enables in-depth analysis of long-term interactive loops in the classroom. Through a bidirectional temporal recurrent neural network, the system can not only record isolated teaching behavior segments, but also identify complex logical loops composed of questioning, answering, and feedback. This advancement transforms the evaluation dimension from fragmented point statistics to structured linear analysis, effectively solving the technical pain point of process data degenerating into isolated noise, and providing a scientific basis for evaluating teachers' heuristic teaching level and classroom control ability.

[0026] 3. This invention relies on an educational logic reasoning intelligent agent, embedding profound educational professional theories into an intelligent evaluation algorithm. Through knowledge graphs and probabilistic graphical models, the system possesses expert-level diagnostic capabilities, providing highly targeted logical feedback for different subjects and teaching stages. This evaluation method, based on semantic logic rather than purely statistical features, significantly enhances the professional depth and interpretability of the evaluation results, making the professional development suggestions generated by the system more advisable.

[0027] 4. The dynamic weighted adaptive evaluation mechanism of this invention breaks the deadlock of fixed evaluation indicators. Through context awareness and reinforcement learning, the system can automatically adjust the weight ratio of various abilities according to the qualifications, subject background, and real-time teaching objectives of the evaluated object. This high degree of flexibility and adaptability enables the system to play a precise guiding role in the evaluation of high-quality and balanced education in developed regions, and also demonstrates excellent scenario adaptability in the diagnosis of teacher capabilities in weak schools under the background of rural revitalization.

[0028] 5. This invention achieves a seamless integration of assessment and development through a closed-loop feedback execution module. The system's low-latency real-time monitoring and intelligent resource recommendation functions transform assessment from a simple "post-event summary" into "pre-event warning" and "in-event guidance." This instant feedback mechanism significantly shortens teachers' self-correction cycle and greatly improves the efficiency of school-based teaching research. Furthermore, the edge computing and cloud-based collaborative architecture ensures large-scale concurrent processing capabilities while rigorously protecting teacher and student privacy, providing secure and reliable technical support for the large-scale implementation of intelligent educational transformation. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall technical solution architecture of the comprehensive evaluation system for teacher education and teaching abilities based on agent-based intelligent agents proposed in this invention; Figure 2 This is a schematic diagram illustrating the core principle framework of multimodal semantic alignment and cross-modal feature mapping in this invention; Figure 3 This is a logical flowchart of the extraction of temporal teaching behavior features and identification of interaction loops in this invention; Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow between edge computing data acquisition and cloud-based intelligent agent reasoning and decision-making in this invention; Detailed Implementation

[0030] Example 1 Please refer to the attached document. Figure 1 This embodiment discloses a comprehensive evaluation system for teachers' educational and teaching abilities based on intelligent agents. This system, through a multi-agent collaborative architecture, achieves deep perception, semantic understanding, logical reasoning, and dynamic evaluation of the entire classroom teaching process. The underlying logic of this system is built upon the integration of edge computing and cloud-based distributed computing. It performs preliminary processing and feature extraction of high-throughput raw data through edge nodes deployed at the teaching site, and utilizes the powerful computing resources of the cloud to execute deep pedagogical logical evolution.

[0031] Combined with appendix Figure 1 The system consists of a multi-source heterogeneous data acquisition module, a multimodal semantic alignment agent, a time-series teaching behavior modeling agent, an educational logic reasoning agent, a dynamic weight adaptive evaluation agent, and a closed-loop feedback execution module.

[0032] The multi-source heterogeneous data acquisition module is deployed in the classroom, achieving comprehensive coverage of the physical teaching space through a high-precision sensor array. This module includes multiple high-resolution video acquisition units responsible for capturing teachers' body movements, facial expressions, and motion trajectories. The video sampling rate is set to 60 frames per second to ensure that high-frequency signals such as micro-expressions are not lost. Simultaneously, the module integrates an array-type microphone pickup unit, employing beamforming technology to directionally enhance the teacher's voice, effectively suppressing background noise in the classroom environment. In addition, the multi-source heterogeneous data acquisition module also includes a teaching material capture unit, a blackboard visual monitoring unit, and a student feedback receiving unit. To ensure absolute synchronization of subsequent multimodal data, the module incorporates a high-precision nanosecond-level timestamp generator, injecting a unified clock marker into each frame of image data, audio sampling point, and interaction log record. During the data transmission phase, the multi-source heterogeneous data acquisition module utilizes gigabit Ethernet or 5G mobile communication technology to push encapsulated data packets to edge computing nodes in real time.

[0033] Please refer to the attached document. Figure 2The multimodal semantic alignment agent receives unstructured data streams from the acquisition module. Its core task is to eliminate semantic offsets between modalities and establish a unified feature representation. Internally, the agent runs a cross-modal shared representation space mapping algorithm, which projects visual behavioral features from the video, acoustic features from the audio, and semantic features from the text after speech-to-text processing into a 512-dimensional high-dimensional feature vector space.

[0034] In a multimodal semantic alignment agent, the system identifies the association strength of different modalities within the same time slice by performing mutual information maximization computation. The computation principle is as follows:

[0035] The above formula describes the mutual information between two modal variables. Wherein, The sequence of speech and semantic features representing the teacher. The sequence of visual action features representing their synchronization. The joint probability distribution of the two. These represent their respective marginal probability distributions. The multimodal semantic alignment agent maximizes mutual information, enabling the system to accurately identify the gesture guidance or blackboard writing position corresponding to the teacher uttering specific keywords.

[0036] Furthermore, the multimodal semantic alignment agent possesses conflict detection and collaborative enhancement mechanisms. When severe environmental noise in the classroom causes the audio modality confidence level to fall below a preset threshold of 0.6, the agent automatically invokes the lip-reading algorithm and body intention recognition model in the visual modality. By analyzing the frequency and amplitude of the teacher's mouth muscle movements and combining this with the current teaching context, it semantically completes the missing key speech fields. The processed alignment data is encapsulated into a multidimensional semantic association matrix, providing a highly consistent input source for subsequent behavior modeling.

[0037] Please refer to the attached document. Figure 3 A temporal teaching behavior modeling agent is connected to a multimodal semantic alignment agent, aiming to analyze the structured features of classroom teaching from a temporal perspective. This agent employs a bidirectional long short-term memory network structure, combined with a time-segmentation proposal network, to segment a 45-minute classroom teaching stream at multiple scales. The system first identifies millisecond-level micro-behaviors, such as the teacher's eye gaze shifts and momentary nodding; then, it aggregates these micro-behaviors into minute-level interaction sequences.

[0038] The temporal teaching behavior modeling agent incorporates a dynamic time warping algorithm, enabling it to adapt to differences in speaking speed and rhythm among teachers explaining the same knowledge point. By analyzing the classroom energy distribution function, the agent automatically divides the teaching process into an introduction, knowledge explanation, teacher-student interaction, classroom practice, and summary phase. When identifying interaction loops, the agent accurately captures the complete logical loop of the teacher initiating a heuristic question, the student providing feedback, and the teacher conducting secondary evaluation and follow-up questions. If an interaction loop is broken temporally—for example, if the teacher answers without giving students sufficient time to think after asking a question—the agent marks that time period as an inefficient interaction point. The behavioral sequence output by the agent includes not only the action itself but also information about the stage of that action within the teaching logic chain.

[0039] The educational logic reasoning agent, serving as the system's central brain, is based on a deep educational knowledge graph containing over one million entities and relationships. This graph encompasses classic theories such as Bloom's Taxonomy of Educational Objectives and Gagné's Nine-Step Teaching Method, and establishes specific evaluation logic branches for different subjects. After the sequential teaching behavior modeling agent outputs behavioral sequences, the educational logic reasoning agent maps these physical-level behaviors into educational semantic symbols using a probabilistic graphical model.

[0040] When dealing with specific subjects, the educational logic reasoning agent exhibits highly differentiated logic. In mathematics teaching assessment, this agent focuses on detecting the logical coherence of teachers when deriving formulas. By comparing the evolution path of the teacher's blackboard writing with the logical nodes in the knowledge graph, it assesses whether the teacher follows the cognitive pattern from concrete to abstract. In language arts teaching assessment, the agent analyzes the consistency between the teacher's tone fluctuations and the text's mood using an affective computing model. When negative fluctuations are observed in student feedback data, the educational logic reasoning agent initiates reverse source reasoning. By traversing the teaching behavior records of the previous 300 seconds, it identifies underlying causes such as excessively large knowledge gaps, vague explanations of key concepts, or overly technical teaching language, thereby generating diagnostic intermediate conclusions.

[0041] The dynamically weighted adaptive evaluation agent is responsible for transforming pedagogical reasoning conclusions into a final competency profile. This agent abandons the rigid percentage allocation of traditional evaluations and introduces a context-aware mechanism. Based on input teaching background parameters, including teacher experience, subject type, student difficulty level, and pre-set teaching objectives, the agent dynamically adjusts the weight coefficients of various evaluation indicators.

[0042] The dynamic weighted adaptive evaluation agent evolves its model based on a reinforcement learning framework. Its state space is defined as the current teaching context, and its action space is defined as the weight vector of each indicator. The system continuously learns from evaluation samples from experienced teaching researchers to optimize its internal evaluation strategy function. When evaluating young teachers with less than three years of experience, the system automatically increases the weights of teaching standardization, classroom management ability, and accuracy of language expression to above 0.4. When evaluating senior teachers, the system automatically switches modes, shifting the evaluation focus towards teaching innovation, depth of heuristic teaching, and higher-order thinking guidance. This mechanism ensures the objectivity and developmental nature of the evaluation conclusions.

[0043] The closed-loop feedback execution module transforms the complex assessment data into actionable professional development suggestions that teachers can understand. This module includes a visual data report generation unit and a natural language generation unit. Quantitative scoring is broken down into four main dimensions—teaching fundamentals, teaching organization, teaching interaction, and teaching effectiveness—and 20 sub-dimensions, presented in radar chart format. Simultaneously, the natural language generation unit automatically synthesizes specific textual guidance suggestions based on weaknesses identified in the assessment, such as suggesting teachers pay more attention to students in the back rows at certain times or recommending more thought-provoking questioning methods.

[0044] Furthermore, the closed-loop feedback execution module possesses millisecond-level real-time alert capabilities. During the teaching process, if the system detects that a teacher is violating teaching routines or that the classroom atmosphere is extremely abnormal, the module will send a specific frequency vibration signal as an early warning through the teacher's smart wearable device. The module also reserves an interface to a school-based professional development platform, which can accurately push master teacher micro-lesson videos or relevant academic articles stored in the cloud resource library based on the weakness tags generated in the current assessment, thus achieving a closed loop integrating assessment and teaching research.

[0045] Please refer to the attached document. Figure 4 This system operates on a hybrid topology that integrates edge and cloud computing. Edge nodes in the teaching environment run multi-source heterogeneous data acquisition modules and multimodal semantic alignment agents, utilizing the GPU resources of edge servers to perform high-concurrency image recognition and speech alignment tasks. This ensures extremely low latency while guaranteeing that sensitive audiovisual data remains within the campus network, protecting data privacy and security. Meanwhile, the computationally intensive educational logic reasoning agent and the dynamic weight adaptive evaluation agent, which rely on large-scale pre-trained models, are deployed in the cloud. Refined feature vectors extracted at the edge are transmitted to the cloud via an encrypted tunnel, and the cloud feeds back the processing results to the display terminals at the edge.

[0046] The system's agents communicate using an asynchronous message passing protocol, with each agent encapsulated as an independent microservice container. Agents exchange data through predefined interfaces, operating independently of each other's internal logic. The system also includes a coordination layer agent responsible for monitoring the operational status of each module. If the multimodal semantic alignment agent experiences processing delays due to computing power fluctuations, the coordination layer agent will automatically initiate task priority scheduling, reducing the sampling rate of some non-critical frames to maintain the continuity of the real-time evaluation stream.

[0047] This system also includes a submodule for monitoring teacher psychological load. This submodule quantifies the teacher's psychological stress level in the classroom by analyzing the fundamental frequency stability of the teacher's voice, fatigue characteristics in facial micro-expressions, and the disorder of their movement trajectory in real time. This indicator is received by the dynamically weighted adaptive evaluation agent and used as a correction factor in the calculation. If the system detects that a teacher is under high pressure or excessively fatigued, it will automatically adjust the scoring tolerance of the teacher's emotional expressiveness dimension, reflecting the scientific nature and humanistic care of the evaluation system.

[0048] This system further realizes distributed multi-classroom collaborative evaluation. Deployed within a regional education network, the system can extract common data on teachers' teaching behaviors across different regions and levels of schools. In this scenario, the multi-agent framework performs global optimization, automatically benchmarking against excellent teaching paradigms within the region, and generating macro-analysis reports for education management departments covering dimensions such as subject balance and teacher quality distribution.

[0049] For assessing digital literacy, the system captures teachers' interactions with digital devices such as smart interactive whiteboards and interactive whiteboards, analyzing the frequency, form, and depth of their access to digital resources. The assessment system can identify whether teachers are merely using interactive whiteboards as a substitute for blackboards, or whether they are making deep use of advanced tools such as dynamic geometry software and virtual experiment platforms, thus accurately evaluating teachers' technological integration capabilities in the digital transformation process.

[0050] Example 2 In another application scenario, the agent-based comprehensive evaluation system for teacher education and teaching abilities involved in this invention can be specially configured for remote dual-teacher classrooms. In this environment, the system needs to simultaneously evaluate the teacher at the main lecturer's end and the teacher at the remote auxiliary end. The multi-source heterogeneous data acquisition module is configured in a dual-end synchronous acquisition mode, with the main lecturer acquiring the teacher's knowledge explanation flow and the remote end acquiring the students' reaction flow and the tutor's organizational flow.

[0051] In this embodiment, the multimodal semantic alignment agent adds cross-spatial alignment functionality. It needs to spatiotemporally match the lecturer's pace with the remote students' head-up rate and interaction response rate. The system establishes a cross-regional semantic association matrix to identify the execution time of the lecturer's instructions in the remote classroom.

[0052] In a dual-teacher model, the temporal teaching behavior modeling agent focuses on the coordination between the lead teacher and the assistant teacher. It assesses the level of synergy in the interaction between the two teachers by identifying the duration of the connection between the lead teacher's "pause request" and the assistant teacher's "intervention guidance." The agent employs a conflict detection model based on a temporal convolutional network to identify whether there are conflicts in teaching instructions between the two teachers at the same time.

[0053] In this embodiment, the educational logic reasoning agent incorporates distance education assessment logic. It expands assessment metrics to dimensions such as remote presence and information transmission loss compensation. The agent infers the teacher's attractiveness in a remote environment by analyzing the visual dwell time of remote students watching a large screen. A probabilistic graphical model is used to handle the uncertainty caused by network latency, reconstructing the actual teaching interaction sequence affected by latency through Bayesian inference.

[0054] In a dual-teacher scenario, the dynamically weighted adaptive evaluation agent automatically increases the weight of the collaborative teaching dimension. For tutors, the system focuses on evaluating their ability to monitor student learning and provide individualized guidance; for lecturers, it focuses on evaluating their ability to coordinate teaching pace and control discourse power across different spaces.

[0055] The closed-loop feedback execution module generates reports in the dual-teacher mode that include a collaborative profile of both ends. It can provide feedback to the lead teacher on which remote classroom has lower participation and push targeted organizational strategies to the assistant teachers. At the same time, the real-time reminder function not only covers mobile phones or wearable devices, but also pops up bubble prompts on the lead teacher's live broadcast interface to inform them of the overall fatigue level of the remote students.

[0056] In this embodiment, the edge computing architecture is distributed across two physical domains. Edge nodes in the two classrooms synchronize feature data in real time via a high-speed dedicated education network. The cloud-based intelligent agent is responsible for aggregating data from both ends and performing global logical reasoning. This distributed architecture effectively solves the problem of limited visibility and in-depth evaluation in traditional remote supervision, enabling high-precision assessment of teaching quality across time and space.

[0057] Example 3 In this embodiment, a comprehensive evaluation system for teachers' educational and teaching abilities based on intelligent agents is applied to a large-scale vocational education training scenario. In this scenario, the multi-source heterogeneous data acquisition module not only includes conventional audiovisual sensors but also integrates real-time monitoring of the operation logs of the training equipment. For example, in CNC machine tool training, the system captures videos of the teacher's fine hand movements during demonstrations and the operation sequences of the machine tool control system, achieving comprehensive capture of teaching behaviors.

[0058] The multimodal semantic alignment agent takes on the task of aligning physical actions with digital instructions. It maps the teacher's gestures to the machine tool's trajectory in real time by sharing a representation space across modalities. When the teacher demonstrates key steps, the system can accurately identify whether their actions conform to professional standards and specifications.

[0059] In a practical training environment, the time-series teaching behavior modeling agent focuses on identifying the cyclical structure of "demonstration, practice, and error correction." The agent models the teacher's patrol trajectory between training stations by constructing a behavior recognition model based on hierarchical reinforcement learning. It can identify which student positions the teacher lingers at for too long and whether any potentially unsafe operational steps have been overlooked.

[0060] In this scenario, the educational logic reasoning agent integrates an industry occupational standards library. It automatically aligns teachers' teaching behaviors with national vocational skill level standards. The agent not only evaluates the advancement of teaching methods but also focuses on assessing the degree of emphasis placed on safe operating procedures. If a teacher omits a safety check step during operation, the logic reasoning agent will automatically assign a score of zero to that indicator and trigger a high-priority warning based on the safety production risk assessment matrix.

[0061] The dynamically weighted adaptive evaluation agent dynamically adjusts the weights of the safety dimensions based on the hazard level of the training task. In training involving high-voltage electricity or high-speed rotating equipment, the weight of the safety regulations dimension is automatically increased to 0.7. However, in purely theoretical explanation sessions, the weight is automatically reduced.

[0062] In this embodiment, the closed-loop feedback execution module provides feedback to teachers via augmented reality glasses. During the teacher's rounds, real-time performance evaluation tags for students are automatically overlaid in their field of vision. If the system detects that a student's training progress is significantly behind schedule, a highlighted area will appear in the teacher's field of vision, along with suggested instructions. The evaluation report at the end of the training will automatically include a graph showing the overlap between the teacher's demonstration and the standard movements, providing quantitative data support for improving the skills of vocational school teachers.

[0063] This specialized application, geared towards practical training, fully demonstrates the flexibility of the system's multi-agent architecture. It can be expanded and restructured in a plug-in manner according to different educational formats, thereby meeting the needs of teacher competency assessment across all industries and disciplines.

[0064] In the actual operation of the system, data security and privacy protection are core elements. The multi-source heterogeneous data acquisition module performs facial desensitization and voice feature blurring at the front end, extracting only feature parameters relevant to teaching behavior for uploading. Edge computing nodes employ hardware-level encryption and decryption chips to ensure absolute data security during transmission and storage. Simultaneously, the system establishes a blockchain-based assessment report traceability mechanism; each generated assessment result is recorded on a private blockchain, ensuring that the assessment conclusions are tamper-proof and legally valid.

[0065] In the collaborative process, the multi-agent system employs a distributed negotiation algorithm. When different agents disagree on the same teaching action—for example, the temporal modeling agent believes it's in an interactive phase while the logical reasoning agent believes it's in a lecturing phase—the coordination layer agent uses a weighted voting mechanism based on the confidence scores of each agent to arrive at a final decision. This internal coordination mechanism based on multi-agent game theory significantly enhances the system's robustness in handling complex and ever-changing teaching scenarios.

[0066] Furthermore, the system's scalability is reflected in its open agent interface protocol. Educational and research institutions can develop custom logic reasoning agent plugins or weight adjustment algorithms based on the latest educational theories, and achieve rapid iterative upgrades of evaluation standards by seamlessly integrating them into the system's middleware layer. This loosely coupled architecture allows the invention to continuously evolve alongside the development of educational science.

[0067] In summary, this invention, by constructing a closed-loop system for multi-agent collaborative work, achieves intelligent processing throughout the entire process, from raw data acquisition to deep logical reasoning, and then to dynamic adaptive evaluation and real-time feedback. Through deep integration of edge and cloud computing, it balances the contradictions between real-time processing, data security, and computational depth, providing a solid technical foundation and scientific methodological support for teacher evaluation reform in my country's digital transformation of education.

[0068] The description of this embodiment is intended to enable those skilled in the art to understand and implement the technical solutions of the present invention. Simple substitutions or parameter adjustments made for specific teaching scenarios or technical environments without departing from the core concept of the present invention should be included within the scope of protection of the present invention. All intelligent agents, modules, and units involved in the system can be implemented through hardware, software, or a combination of hardware and software, and their physical deployment is not limited to the specific architecture described in the above embodiments.

[0069] In the advanced evolution of the system, the dynamic weighted adaptive evaluation agent also incorporates a long-term memory network to track teachers' professional growth trajectory over a semester or longer. It not only analyzes the performance of individual lessons but also assesses the speed at which teachers optimize their teaching strategies through cross-lesson comparisons. This long-term evaluation perspective enables the system to identify potential teacher burnout or professional development bottlenecks, providing school administrators with more forward-looking personnel management decision-making suggestions.

[0070] The educational logic reasoning agent can also connect in real time with global databases of educational papers. It extracts the latest pedagogical research findings using natural language processing technology and transforms them into logical reasoning rules. This means that when a new, highly effective teaching method is recognized by the academic community, the system can automatically update its knowledge graph via the cloud, achieving instant global synchronization of assessment standards.

[0071] At the hardware support layer, the system employs heterogeneous computing accelerator cards, improving the efficiency of complex neural network operations by more than 10 times. The visual acquisition unit in the multi-source heterogeneous data acquisition module utilizes a global shutter sensor, completely eliminating motion blur caused by the teacher's rapid movement. The audio acquisition unit achieves sub-millisecond echo cancellation and environmental noise reduction through an integrated digital signal processor. These extreme optimizations of the underlying hardware provide high-quality raw materials for the algorithm execution of the upper-level intelligent agents.

[0072] The interactive design of the closed-loop feedback execution module follows the principles of cognitive psychology. Its generated visual interface employs a hierarchical information display strategy. The first screen displays only the core competency scores and risk warnings, allowing users to drill down to details of each minute of teaching behavior and original video clips. This design avoids the psychological pressure on teachers caused by information overload, making the assessment system more like a personal digital tutor than a monitoring tool.

[0073] The distributed collaborative evaluation mechanism employs federated learning technology during its implementation. Raw data between different schools is not shared; only gradient information for model updates is exchanged. This allows the system to build a high-precision evaluation benchmark model covering the entire region while protecting the data sovereignty of each school. This technological approach effectively solves the problem of siloed educational data and promotes balanced regional educational development.

[0074] Ultimately, this agent-based comprehensive evaluation system for teachers' teaching abilities transforms previously vague and subjective teaching assessments into precise and objective data insights. It captures fleeting moments of teaching inspiration and interactive sparks in the classroom, solidifying them into analyzable and replicable teaching models. This not only enhances individual teachers' capabilities but also lays the necessary technical foundation for building a data-driven, intelligent education ecosystem.

[0075] To ensure system stability in extreme environments, the coordination layer agent incorporates a degradation strategy. When network bandwidth falls below 10 megabits per second, the system automatically cuts off high-definition video streaming and switches to a lightweight feature stream transmission mode based on skeleton point information. In this mode, the multimodal semantic alignment agent relies heavily on speech semantics and human skeleton vectors for reasoning. This flexible switching capability ensures that the system can still stably provide core assessment services in remote rural schools or campuses with complex network environments.

[0076] To verify the scientific validity of the evaluation indicators, the system introduces a goodness-of-fit verification mechanism between expert scoring and agent scoring. After the system completes a certain number of automatic evaluations, 1% of the samples are randomly selected for double-blind evaluation by a panel of human experts. The dynamically weighted adaptive evaluation agent fine-tunes its internal weighting strategy function by comparing the residuals of the two evaluations and using an error backpropagation algorithm. Through this closed-loop continuous optimization through human-machine collaboration, the system's evaluation authority is continuously strengthened.

[0077] The teacher psychological workload monitoring submodule has demonstrated significant clinical value during operation. Through statistical analysis of numerous teaching cases, the system can identify specific teaching scenarios that lead to sudden increases in teacher stress, such as handling unexpected classroom discipline issues or responding to questions on challenging knowledge points. The closed-loop feedback execution module uses this data to customize stress management suggestions for teachers and provides reasonable rest and professional development recommendations within the scheduling system, truly achieving full-lifecycle care for teachers' professional development.

[0078] The system proposed in this invention is not merely an assessment tool, but a paradigm shift in educational evaluation. It moves teaching evaluation from outcome-oriented to process-oriented, from a single dimension to a multi-modal perspective, and from qualitative judgment to quantitative analysis. Through the refined division of labor and efficient collaboration among multiple agents, this system successfully solves the analytical challenges of the black-box system of classroom teaching, opening up a completely new path for the scientific management of education and teaching. In future applications, with the improvement of computing power and the iteration of algorithms, this system will demonstrate even stronger adaptability to different scenarios and greater educational insight.

[0079] The comprehensive evaluation system for teachers' teaching abilities provided by this invention integrates advanced artificial intelligence technology, big data analysis technology, and educational theory to construct a comprehensive, multi-level evaluation framework. This system not only provides individual teachers with personalized improvement plans but also offers schools a holistic profile of their teaching staff and provides precise decision-making support for education authorities. Its innovative multi-agent architecture mechanism ensures extremely high accuracy and robustness when processing complex, unstructured teaching data.

[0080] In the specific algorithm implementation of multimodal data alignment, the system also employs contrastive learning technology. By constructing positive and negative sample pairs, it further narrows the feature distance of the same teaching behavior in different modalities and widens the distance between different behaviors. This improvement enables the system to achieve pixel-level accuracy in recognizing teachers' subtle movements; even slight gesture changes can be accurately captured and assigned semantic labels.

[0081] In the temporal modeling process, the system introduces an attention mechanism, enabling the agent to automatically focus on key moments in classroom teaching, such as the instant of knowledge point transition or the moment when a student has a question. By giving high weight to these key moments, the relevance of the evaluation results is significantly improved.

[0082] For logical reasoning in education, the system also establishes a dynamically evolving case library. Whenever the system identifies a highly representative successful teaching case or a typical teaching failure case, the case and its underlying logical chain are automatically anonymized and stored in the case library. In subsequent work, the logical reasoning agent will retrieve similar cases to assist in its logical judgment, enhancing the persuasiveness of the evaluation conclusions.

[0083] The closed-loop feedback execution module also integrates speech synthesis technology, providing teachers with voice-based assessment summaries, allowing them to receive professional advice during fragmented time such as commutes. Simultaneously, the system supports deep integration with various mainstream teaching software and office platforms, enabling teachers to directly access assessment feedback within their frequently used work interfaces, significantly lowering the technological barrier to entry.

[0084] In terms of system deployment and operation, the system adopts a containerized cluster management solution, supporting one-click deployment and automated scaling. Whether in a small-scale application in a single school or a large-scale deployment at the provincial or municipal level, the system can ensure the efficient completion of assessment tasks by dynamically adjusting resource allocation. The system backend also has comprehensive log auditing and fault self-diagnosis functions, capable of locating system anomalies and triggering automatic repair scripts within milliseconds.

[0085] This invention describes the system's core technologies and various application scenarios, aiming to demonstrate its broad applicability and technological leadership. In practical engineering implementation, the modules can be rationally combined and tailored according to specific cost budgets and performance requirements. For example, in basic application scenarios, only data acquisition, semantic alignment, and basic evaluation modules can be retained; while in advanced scientific research scenarios, the complete educational reasoning and dynamic weight evolution functions can be enabled. Regardless of the configuration, the multi-agent collaborative evaluation concept advocated by this invention can realize its core value.

[0086] Through the detailed description of the above implementation methods, it is evident that this invention possesses unique advantages in solving core technical problems such as modal alignment, long-term modeling, professional logical reasoning, and adaptive weight adjustment. It not only achieves breakthroughs at the technical level but also aligns with the real needs of educational practice at the application level, possessing extremely high value for widespread application and significant social benefits. In the future, with the emergence of new forms such as the educational metaverse and human-computer collaborative teaching, this system will continue to absorb more cutting-edge technologies through its open intelligent agent architecture, continuously leading the intelligent wave in the field of educational evaluation.

[0087] Furthermore, when assessing teachers' digital literacy, this invention not only monitors device usage frequency but also deeply analyzes how teachers utilize digital tools for differentiated instruction. For example, the system can identify whether teachers dynamically adjust their explanation pace based on students' real-time online practice data. This assessment of the depth of technology integration better reflects teachers' true teaching level in the digital age. By collecting teachers' records of reconstructing digital resources, guiding strategies for online discussions, and feedback on the grading of digital assignments, the system constructs a panoramic model of teachers' digital literacy, providing precise guidance for teachers' digital transformation.

[0088] At the system's global management layer, an ethics monitoring agent is also present. This agent is responsible for overseeing all assessment actions to ensure they comply with educational ethical standards, guaranteeing that the algorithm does not produce gender, regional, or subject-specific discrimination. The ethics monitoring agent performs real-time statistical analysis of the distribution of assessment results. If it detects an abnormal deviation in a particular indicator within a specific group, it will immediately freeze the scoring function for that indicator and submit it for manual review. This design ensures that the application of artificial intelligence in educational assessment remains safe, fair, and controllable, reflecting the concept of responsible AI development.

[0089] At the data visualization level, the closed-loop feedback execution module employs immersive data presentation technology. It recreates the classroom scene using 3D modeling technology and overlays assessment data onto the virtual environment in the form of heatmaps, path maps, and other formats. Teachers can review their teaching process using virtual reality devices, intuitively assessing the rationality of their classroom layout and the balance of their questioning. This intuitive visual feedback is more impactful than traditional textual feedback and effectively motivates teachers to improve themselves.

[0090] Finally, the system's multi-agent architecture also supports cross-platform and cross-terminal collaborative work. Teachers can receive briefings on their mobile phones, view detailed diagnoses on tablets, participate in school-based professional development on desktops, and receive real-time alerts on smart blackboards. This seamless cross-screen experience allows the assessment system to be deeply embedded in teachers' daily workflows, truly achieving accompanying and seamless intelligent evaluation, and providing a powerful technological engine for building a lifelong learning-oriented teacher professional development system.

Claims

1. A comprehensive evaluation system for teachers' educational and teaching abilities based on intelligent agents, characterized in that, include: A multi-source heterogeneous data acquisition module is used for real-time accompanying data capture in a classroom teaching environment; The multi-source heterogeneous data acquisition module acquires multi-dimensional raw data, including video stream data, audio stream data, teaching courseware text image data, blackboard dynamic visual data, and student feedback data, through a high-precision sensor array. The module utilizes a built-in high-precision timestamp generator to inject a unified nanosecond-level synchronization identifier into each frame of video image, audio sampling point, and interaction log to establish a physical layer synchronization foundation. The multi-modal semantic alignment agent projects unstructured video behavioral features, audio acoustic features, and text semantic features into a unified high-dimensional feature vector space by executing a cross-modal shared representation space mapping algorithm. The multimodal semantic alignment agent identifies the semantic association strength between teacher's verbal expression, body language, and blackboard writing by calculating the mutual information maximization function between different modal variables, and constructs a multi-dimensional semantic association matrix. The temporal teaching behavior modeling agent uses a bidirectional temporal recurrent neural network structure combined with a time-segmentation proposal network to perform multi-scale segmentation and behavior modeling of the complete classroom teaching process. This agent identifies complete interactive loops containing questioning, answering, and feedback by constructing a long short-term memory network, and uses a dynamic time warping algorithm to adapt and model the differences in teaching rhythms among different teachers and subjects. The educational logic reasoning agent maps the identified behavioral sequences to educational language based on a built-in educational knowledge graph. The system includes: a semantic symbol; an educational logic reasoning agent that performs logical reasoning tailored to different subject characteristics using a probabilistic graphical model, and performs reverse tracing reasoning based on student feedback data to locate logical anomalies in pre-teaching behaviors and generate diagnostic intermediate assessment conclusions; a dynamic weighted adaptive assessment agent that introduces a context-aware mechanism to dynamically adjust the priority weights of each assessment indicator based on the current teaching task objectives, teacher professional background, and student learning foundation using a reinforcement learning framework to generate a multi-dimensional teacher competency profile; and a closed-loop feedback execution module that transforms assessment results into visualized data reports and improvement strategy suggestions, and provides real-time reminders through a low-latency monitoring interface, while also matching and pushing professional development resources from a cloud resource library based on the assessment conclusions.

2. The comprehensive evaluation system for teacher education and teaching ability based on intelligent agents according to claim 1, characterized in that, The multimodal semantic alignment agent is also used to perform conflict detection and collaborative enhancement processing: when the confidence level of the audio modality is lower than a preset threshold due to environmental noise interference, the multimodal semantic alignment agent calls the lip reading recognition features and body intention features in the visual modality to restore and correct the lost speech semantics; the mutual information maximization function quantifies the correlation strength of different modalities in the same time slice by calculating the expected value of the log ratio of the joint probability distribution and the marginal probability distribution of each modality variable.

3. The comprehensive evaluation system for teachers' educational and teaching abilities based on intelligent agents according to claim 1, characterized in that, The time-series teaching behavior modeling agent is also used to automatically label the teaching process as an introduction, explanation, practice, and summary stage by analyzing the fluctuations in classroom energy distribution function and interaction frequency. The time-series teaching behavior modeling agent is also used to automatically identify transitional phrases and logical connection points when switching teaching stages based on historical evaluation benchmarks, thereby quantitatively modeling the teacher's ability at the classroom structure organization level.

4. The comprehensive evaluation system for teacher education and teaching ability based on intelligent agents according to claim 1, characterized in that, The educational logic reasoning agent is built into an educational knowledge graph that covers Bloom's Taxonomy of Educational Objectives and Gagné's Taxonomy of Learning Outcomes. When performing logical reasoning for different subject characteristics, the educational logic reasoning agent focuses on testing the logical coherence of formula derivation and the process of abstract thinking for mathematics, and analyzes the fit between intonation and textual meaning through an affective computing model for language arts.

5. A comprehensive evaluation system for teachers' educational and teaching abilities based on intelligent agents according to claim 1, characterized in that, In the reinforcement learning framework adopted by the dynamic weight adaptive evaluation agent, the state space is defined as the parameters of the current teaching situation, and the action space is defined as the weight vector of each evaluation indicator. The dynamic weight adaptive evaluation agent sets different evaluation weight strategies according to the teacher's teaching experience: for young teachers with less than 3 years of service, the weight of teaching standardization and classroom management ability is automatically increased to above 0.4; for senior teachers, the evaluation focus is automatically switched to teaching innovation and higher-order thinking guidance ability.

6. The comprehensive evaluation system for teachers' educational and teaching abilities based on intelligent agents according to claim 1, characterized in that, The closed-loop feedback execution module includes a visualization data report generation unit and a natural language generation unit. The visualization data report generation unit is used to generate quantitative scores that include four main dimensions (teaching basic skills, teaching organization, teaching interaction, and teaching effectiveness) and 20 sub-dimensions, and present them in a radar chart. The closed-loop feedback execution module is also used to send a vibration warning signal of a specific frequency through the smart wearable device worn by the teacher when the system detects that the teacher has violated teaching routines or that the classroom atmosphere is abnormal.

7. The comprehensive evaluation system for teacher education and teaching ability based on intelligent agents according to claim 1, characterized in that, Running on a hybrid architecture of edge computing and cloud collaboration: the multi-source heterogeneous data acquisition module and the multimodal semantic alignment agent are deployed on the school's local edge computing nodes to perform high-concurrency image recognition, speech alignment and feature extraction tasks; The educational logic reasoning agent and the dynamic weight adaptive evaluation agent are deployed on a cloud server to perform deep logic analysis using a large-scale pre-trained model; the edge computing nodes and the cloud server communicate asynchronously via encrypted tunnels to exchange refined feature vectors.

8. The comprehensive evaluation system for teachers' educational and teaching abilities based on intelligent agents according to claim 1, characterized in that, It also includes a teacher psychological load monitoring submodule, which is used to quantify the teacher's psychological stress value in classroom teaching by analyzing the teacher's voice fundamental frequency stability, fatigue characteristics in facial micro-expressions, and the smoothness of movement trajectory in the classroom; the psychological stress value is input as a correction factor into the dynamic weight adaptive evaluation agent, which is used to automatically adjust the scoring tolerance of the emotional expression dimension when the teacher is under high pressure or excessive fatigue.

9. A comprehensive evaluation system for teachers' educational and teaching abilities based on intelligent agents according to claim 1, characterized in that, It also includes a digital literacy assessment unit, which is used to analyze the frequency of teachers' use of digital teaching resources, the form of interaction, and the depth of integration by capturing the interaction trajectory between teachers and smart teaching devices; the digital literacy assessment unit is used to identify teachers' advanced application behaviors of dynamic geometry software and virtual experimental platforms, and, in combination with the operation logs of teaching software, to accurately evaluate teachers' technical integration capabilities in the process of digital transformation.

10. A comprehensive evaluation system for teacher education and teaching ability based on intelligent agents according to claim 1, characterized in that, It also includes a distributed multi-classroom collaborative evaluation mechanism, which is used to incorporate multiple geographically dispersed teaching venues into a unified evaluation network; the system performs global coordination through a multi-agent framework, automatically extracts the common characteristics and differentiated performance of teachers' teaching abilities in different regions, and uses federated learning technology to construct a benchmark model of teaching ability covering the entire region while protecting the data privacy of each school, so as to generate a macro-level teacher quality analysis report.

Images