Teacher evaluation method and system based on multi-modal data

Abstract

The invention belongs to the technical field of data analysis, and discloses a teacher evaluation method and system based on multi-modal data. The method comprises the following steps: based on a teacher professional development evaluation platform, according to collected real-time multi-modal data of a teacher, carrying out teacher evaluation to obtain a real-time teacher score and a predicted teacher score; according to the real-time teacher score and the predicted teacher score, personalized development suggestions are generated, and real-time personalized development suggestions are obtained and visually displayed. According to the method, the problems of single evaluation dimension, non-uniform evaluation standards and low efficiency in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the field of data analysis technology, and specifically relates to a teacher evaluation method and system based on multimodal data. Background Art

[0002] Teacher evaluation is an essential process for education departments and schools to assess new teachers, promote key teachers, monitor regional teaching staff dynamics, and assist in the review of teacher title promotions. Accurate, comprehensive, and comprehensive teacher evaluations have become a key research area in the education industry.

[0003] The existing technology has many defects, including:

[0004] 1) Single evaluation dimension: Traditional evaluation methods rely too heavily on subjective qualitative judgments, such as leadership ratings of lectures, peer reviews, and student satisfaction questionnaires. These evaluations often focus on only a portion of a teacher's performance, such as verbal expression in class, teaching attitude, or student feedback, while ignoring important information such as pre-class preparation, after-class tutoring, participation in teaching and research activities, technology application skills, satisfaction of students' individual needs, and emotional intelligence. This makes it difficult to form a comprehensive and three-dimensional understanding of a teacher's professional competence.

[0005] 2) Inconsistent evaluation criteria: When relying on manual evaluation, the evaluator's personal experience, preferences, emotions, and even the relationship with the evaluated individual may affect the evaluation results, leading to inconsistent and unfair evaluation criteria. Different evaluators may give completely different evaluations of the same teacher, and the same evaluator may give significantly different evaluations of the same teacher at different times.

[0006] 3) Inefficiency: As the number of teachers increases, the data used for evaluation increases exponentially. The existing manual evaluation method results in a huge workload for evaluators and the efficiency cannot meet the requirements. Summary of the Invention

[0007] In order to solve the problems of single evaluation dimension, inconsistent evaluation standards and low efficiency in the existing technology, the purpose of the present invention is to provide a teacher evaluation method and system based on multimodal data.

[0008] The technical solution adopted in the present invention is:

[0009] A teacher evaluation method based on multimodal data includes the following steps:

[0010] Based on the teacher professional development evaluation platform, teachers are evaluated based on the collected real-time multimodal data of teachers, and real-time teacher scores and predicted teacher scores are obtained;

[0011] Based on real-time teacher ratings and predicted teacher ratings, personalized development suggestions are generated, real-time personalized development suggestions are obtained, and visually displayed.

[0012] Furthermore, the teacher professional development evaluation platform is equipped with a multi-dimensional evaluation indicator mathematical model, an intelligent evaluation model, a scoring prediction model and a professional development suggestion model.

[0013] Furthermore, the multi-dimensional evaluation index mathematical model includes a secondary index aggregation mathematical sub-model, a primary index aggregation mathematical sub-model, a comprehensive evaluation score mathematical sub-model, and a value-added evaluation mathematical sub-model;

[0014] The intelligent evaluation model includes a multimodal data feature extraction module, a multi-task learning module, a feature fusion module, and an intelligent evaluation module;

[0015] The rating prediction model includes a time series feature extraction module and a rating prediction module;

[0016] The professional development suggestion model includes an experience replay pool, an agent, and a reward function.

[0017] Furthermore, based on the teacher professional development evaluation platform, teacher evaluation is performed according to the collected real-time multimodal data of teachers to obtain real-time teacher scores and predicted teacher scores, including the following steps:

[0018] Collect teachers' real-time multimodal data, pre-process the real-time multimodal data, and store the pre-processed real-time multimodal data in the teacher professional development evaluation platform;

[0019] Based on the teacher professional development evaluation platform, an intelligent evaluation model is used to evaluate teachers based on pre-processed real-time multimodal data, obtaining real-time three-level indicator teacher scores;

[0020] Based on the real-time three-level indicator teacher ratings, a multi-dimensional evaluation indicator mathematical model is used to generate corresponding real-time teacher ratings;

[0021] Based on the real-time teacher ratings, the rating prediction model is used to make rating predictions and obtain the predicted teacher ratings.

[0022] Furthermore, based on the teacher professional development evaluation platform, an intelligent evaluation model is used to evaluate teachers based on the pre-processed real-time multimodal data, and obtain real-time three-level indicator teacher evaluation, including the following steps:

[0023] Based on the teacher professional development evaluation platform, the multimodal data feature extraction module of the intelligent evaluation model is used to extract the real-time multimodal data features of the pre-processed real-time multimodal data;

[0024] According to the preset attention weight value, the feature fusion module of the intelligent evaluation model is used to perform weighted fusion on the real-time multimodal data features to obtain real-time weighted fusion features;

[0025] According to the real-time weighted fusion features, the intelligent evaluation module of the intelligent evaluation model is used to evaluate teachers and obtain real-time three-level indicator teacher evaluations.

[0026] Furthermore, based on the real-time three-level indicator teacher ratings, a multi-dimensional evaluation indicator mathematical model is used to generate corresponding real-time teacher ratings, including the following steps:

[0027] Input the real-time third-level indicator teacher ratings into the second-level indicator aggregation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time second-level indicator teacher ratings;

[0028] Input the real-time second-level indicator teacher rating into the first-level indicator aggregation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time first-level indicator teacher rating;

[0029] Input the real-time first-level indicator teacher rating into the comprehensive evaluation score mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time comprehensive teacher rating;

[0030] Input the real-time first-level indicator teacher rating and the real-time comprehensive teacher rating into the value-added evaluation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time development increment;

[0031] Integrate the real-time first-level indicator teacher rating, real-time comprehensive teacher rating and real-time development increment to obtain the corresponding real-time teacher rating.

[0032] Furthermore, based on the real-time teacher ratings, a rating prediction model is used to perform rating prediction to obtain predicted teacher ratings, including the following steps:

[0033] Use the time series feature extraction module of the rating prediction model to extract the real-time time series features of the real-time first-level indicator teacher rating in the real-time teacher rating;

[0034] According to the real-time time series characteristics, the score prediction module of the score prediction model is used to perform score prediction and obtain the predicted first-level indicator teacher score;

[0035] Input the predicted first-level indicator teacher score into the comprehensive evaluation score mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the predicted comprehensive teacher score;

[0036] Input the predicted first-level indicator teacher score and the predicted comprehensive teacher score into the value-added evaluation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the predicted development increment;

[0037] Integrate the predicted first-level indicator teacher rating, the predicted comprehensive teacher rating, and the predicted development increment to obtain the corresponding predicted teacher rating.

[0038] Furthermore, based on the real-time teacher ratings and the predicted teacher ratings, personalized development suggestions are generated, and real-time personalized development suggestions are obtained and visualized, including the following steps:

[0039] If the real-time first-level indicator teacher score in the real-time teacher score is lower than the score threshold, then proceed to the next step; otherwise, continue with the teacher evaluation process;

[0040] Input real-time teacher ratings and predicted teacher ratings into the professional development recommendation model of the teacher professional development evaluation platform;

[0041] According to the real-time first-level indicator teacher rating in the real-time teacher rating, the state space of the intelligent agent of the professional development suggestion model is updated to obtain an updated state space;

[0042] updating the reward function of the professional development recommendation model according to the predicted development increment in the predicted teacher ratings to obtain an updated reward function;

[0043] Based on the updated action space and the updated reward function, the intelligent agent of the professional development suggestion model is used to generate personalized development suggestions and obtain real-time personalized development suggestions;

[0044] Store and visualize teachers' real-time teacher ratings, predicted teacher ratings, and real-time personalized development suggestions.

[0045] A teacher evaluation system based on multimodal data is used to implement a teacher evaluation method. The system includes a teacher professional development evaluation platform and an external terminal. The teacher professional development evaluation platform is communicatively connected to the external terminal.

[0046] The teacher professional development evaluation platform is built based on a layered distributed architecture, which includes an infrastructure layer, a data middle platform layer, an application service layer, and a user interaction layer. The user interaction layer is connected to external terminals for communication.

[0047] Furthermore, the infrastructure layer includes a distributed computing module and a storage module;

[0048] The data middle platform layer includes the data governance module, knowledge graph module, and user portrait module;

[0049] The application service layer includes intelligent evaluation service module and decision support service module;

[0050] The user interaction layer includes the teacher client and the management client.

[0051] The beneficial effects of the present invention are:

[0052] The present invention provides a teacher evaluation method and system based on multimodal data. The multi-dimensional evaluation index mathematical model constructed provides a city teacher professional development index system and constructs a three-dimensional evaluation framework including several first-level indicators, several second-level indicators and several third-level indicators. It can break through the limitations of traditional single evaluation dimension, reflect the real teaching status and professional performance of teachers from a more comprehensive and multi-angle perspective, reduce subjective bias, and improve the objectivity and comprehensiveness of evaluation; based on educational measurement theory, it integrates advanced technologies such as deep learning, knowledge graphs, and federated learning to construct a three-dimensional support system of "scientific indicators, intelligent evaluation, and scenario-based application"; with "multimodal data" as the core input, and processed through professional models such as intelligent evaluation models and scoring prediction models, it effectively integrates and deeply mines various data generated in the classroom (including traditional methods The system can collect and analyze the unstructured data ignored by teachers, transform these data into valuable evaluation information and decision-making basis, and reflect the advantages of technology-driven development. Based on real-time scoring and predictive scoring, the professional development suggestion model generates "real-time personalized development suggestions". These suggestions can be closely combined with the current specific performance of teachers and future development trends, and provide more targeted and feasible improvement directions and development paths, effectively solving the problem of "one size fits all", and better meeting the professional growth needs of individual teachers. Based on the teacher professional development evaluation platform, it realizes automated and intelligent data collection, intelligent evaluation and personalized suggestions, reduces the workload of evaluators, and realizes multi-party linkage through teacher clients and management clients, building an evaluation ecosystem with data interoperability, functional collaboration, and covering teachers and managers, thereby improving the overall efficiency and coordination of evaluation work.

[0053] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flowchart of the teacher evaluation method based on multimodal data in the present invention.

[0055] Figure 2 It is a structural block diagram of the teacher evaluation system based on multimodal data in the present invention. DETAILED DESCRIPTION

[0056] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0057] Example 1:

[0058] like Figure 1 As shown, this embodiment provides a teacher evaluation method based on multimodal data, including the following steps:

[0059] S1: Based on the teacher professional development evaluation platform, teachers are evaluated according to the collected real-time multimodal data of teachers to obtain real-time teacher scores and predicted teacher scores;

[0060] The teacher professional development evaluation platform is equipped with a multi-dimensional evaluation indicator mathematical model, an intelligent evaluation model, a scoring prediction model, and a professional development suggestion model;

[0061] The intelligent evaluation model, score prediction model, and professional development recommendation model use Federated Learning technology. Without sharing original data, this technology achieves cross-school data collaborative modeling through local model training and parameter aggregation. The communication protocol complies with the "Information Security Technology Personal Information Security Specification" (GB / T 35273).

[0062] The data transmission of the teacher professional development evaluation platform adopts TLS1.3 encryption, and the data storage implements hierarchical access control. Teacher data access requires two-factor authentication (username + dynamic token);

[0063] The multi-dimensional evaluation index mathematical model includes a secondary index aggregation mathematical sub-model, a primary index aggregation mathematical sub-model, a comprehensive evaluation score mathematical sub-model, and a value-added evaluation mathematical sub-model;

[0064] The formula of the secondary indicator aggregation mathematical sub-model is:

[0065]

[0066] Where, S(B i ) is the teacher score of the secondary indicator; w ij For example, the weight of the third-level indicator "teacher ethics" in the second-level indicator "professional quality and teacher ethics practice" is 60%; S(C ij ) is the teacher score of the third-level indicator; i is the indicator quantity of the second-level indicator; j is the indicator quantity of the third-level indicator; m is the total number of the third-level indicators;

[0067] The three-level indicator teacher rating is the value output by the intelligent evaluation model, which is quantified using the Likert 5-level scale, with 1-5 points corresponding to "unqualified" to "excellent". Taking "teacher ethics" as an example, it includes three three-level indicators: "ideological and moral character", "health ethics" and "personal cultivation", with weights of 33.39%, 33.39%6 and 33.49% respectively. The score result is: S (teacher ethics) = 0.333*S (ideological and moral character) + 0.333*S (professional ethics) + 0.334*S (personal cultivation);

[0068] The formula of the first-level indicator aggregation mathematical sub-model is:

[0069]

[0070] Where, S(A k ) is the first-level indicator teacher score; w kl is the weight of the secondary indicator; S(B kl ) is the teacher score of the secondary indicator; k is the indicator quantity of the primary indicator; l is the indicator quantity of the secondary indicator; n is the total number of secondary indicators;

[0071] For example, the teaching and practice ability (weight 34.9%) is a weighted synthesis of secondary indicators such as classroom teaching ability (23.5%), class management and education ability (23.59%), teaching, scientific research and innovation ability (17.6%), collaborative communication ability (17.6%), and curriculum construction ability (17.6%). The score result is: S (teaching and practice ability) = 0.235*S (classroom teaching ability) + 0.235*S (class management and education ability) + 0.176*S (teaching, scientific research and innovation ability) + 0.176*S (collaborative communication ability) + 0.176*S (curriculum construction ability);

[0072] The formula of the comprehensive evaluation score mathematical sub-model is:

[0073]

[0074] Where, T is the comprehensive evaluation score; w k The weight of the first-level indicator for comprehensive evaluation;

[0075] The weights of the first-level indicators are: professional quality and teacher ethics (20%), utilitarianism and academic foundation (16%), teaching and practical ability (34%), lifelong learning and self-improvement (15%), and digital literacy and intelligent education (15%). The weights will be dynamically adjusted through the Delphi method combined with education policy guidance. For example, during the digital transformation stage of education, the weight of digital literacy and intelligent education can be temporarily increased to 20%.

[0076] The formula of the mathematical sub-model of value-added evaluation is:

[0077]

[0078] Where ΔT is the development increment; T 末 ,T 初 The comprehensive evaluation score at the end and beginning of the evaluation cycle; The average improvement value of the first-level indicator teacher rating of similar teachers in the region is eliminated through the mixed effect model to eliminate the interference of background factors such as professional title and teaching experience, highlighting the effectiveness of individual development; k ' is the weight of the first-level indicator of value-added evaluation; the mathematical sub-model of value-added evaluation is used to calculate the progress of teachers within the evaluation cycle (2-3 years);

[0079] Based on the adaptive weight optimization mechanism, the indicator weights are dynamically adjusted according to the teacher's career stage (novice / backbone / excellent) and regional education policy orientation. The formula of the adaptive weight optimization mechanism is:

[0080]

[0081] Where, is the weight of the k'th indicator at time step t', t'-1; t', t'-1 are time step indicators; k' is the indicator indicator; η is the learning rate (initial value 0.01, decaying with training). The weight adjustment strategy is dynamically updated through meta-learning, and the model parameters are calibrated once a quarter. The loss function L is the weight gradient; for The L2 norm of

[0082] The intelligent evaluation model includes a multimodal data feature extraction module, a multi-task learning module, a feature fusion module, and an intelligent evaluation module;

[0083] The multimodal data feature extraction module includes a first input layer, a structured data feature extraction layer, and a text data feature extraction layer. The first input layer is used to receive preprocessed multimodal data. The structured data feature extraction layer uses a convolutional neural network (CNN) to extract structured data features of structured data in the multimodal data. The text data feature extraction layer uses a bidirectional encoder representation from transformers (BERT) model to extract text data features of text data in the multimodal data.

[0084] The multi-task learning module includes a shared feature extraction layer and a task-specific layer. The shared feature extraction layer is used to extract shared features of the multimodal data feature extraction module. The task-specific layer is equipped with a multi-task model to simultaneously optimize sub-tasks such as teacher ethics evaluation, teaching ability diagnosis, and scientific research effectiveness analysis, thereby improving the accuracy of the intelligent evaluation model.

[0085] The formula of the multi-task model is:

[0086]

[0087] Where L is the loss function of the multi-task model; λ k is the subtask weight (dynamic adjustment range 0.1-0.3); L kis the subtask loss function (e.g., cross entropy for teacher ethics evaluation and mean square error for teaching ability diagnosis); γ is the L2 regularization coefficient (value is 0.001); θ is the multi-task model parameter;

[0088] The feature fusion module introduces a self-attention mechanism to dynamically adjust the attention of different features;

[0089] The formula of the feature fusion module is:

[0090]

[0091] In the formula, Attention(Q,K,V) is the weighted fusion feature; softmax(*) is the activation function; Q, K, V are the query vector, key vector, and value vector, respectively corresponding to the teacher's moral performance, teaching effect, training record and other features; T is the transposition symbol; d k is the key vector dimension;

[0092] The rating prediction model includes a time series feature extraction module and a rating prediction module;

[0093] The temporal feature extraction module uses a long short-term memory network (LSTM) to capture the temporal laws of teacher development. The input is the first-level indicator teacher rating sequence X. t =[x t1 ,x t1 ,...,x t5 ](x t1 ,x t1 ,...,x t5 Corresponding to the monthly scores of the five first-level indicators), the hidden layer state update formula is:

[0094] f t =σ(W f ·[X t ,h t-1 ]+b f )

[0095] i t =σ(W i ·[X t ,h t-1 ]+b i )

[0096]

[0097] o t =σ(W o ·[X t ,h t-1 ]+b o )

[0098] h t =o t ⊙tanh(C t )

[0099] Where, X t is the first-level indicator teacher rating sequence; h t-1 is the hidden layer state of the previous step; W f ,W i ,W C ,W o is the weight matrix of the forget gate, input gate, cell state candidate value and output gate; b f ,b i ,b C ,b o is the bias term of the forget gate, input gate, cell state candidate value and output gate; σ(*) is the sigmoid activation function, which is used to control the degree of gate opening; tanh(*) is the hyperbolic tangent activation function, which is used to adjust the value of the cell state; ⊙ is the element-wise multiplication operation; f t is the output of the forget gate, which determines the cell state C at the previous moment t-1 The degree of preservation; C t-1 is the cell state at the previous moment; i t Is the output of the input gate, which determines the candidate value of the new cell state The degree to which it is added to the current cell state; is the candidate value of the new cell state; C t is the cell state at the current moment; o t is the output of the output gate, which determines the current cell state C t The degree of output; h t is the hidden layer state at the current moment;

[0100] The rating prediction module generates predictions for the next three months through a fully connected layer to assess teacher development potential;

[0101] The professional development suggestion model includes an experience replay pool, an agent, and a reward function;

[0102] The formula of the reward function is:

[0103] R(s,a)=α·ΔT 预测 +β·cos(s,a)+γ·D(a)

[0104] Where R(s,a) is the reward function; ΔT 预测 is the predicted development increment; cos(s,a) is the capability-resource cosine similarity; D(a) is the duration of action a; α, β, γ are the first weight parameter, the second weight parameter, and the third weight parameter; s is the state parameter; a is the action parameter;

[0105] The experience replay pool is used to store and replay past "experiences." Like a database, it records the actions taken by the agent (individual or system) during its professional development (such as learning a skill, participating in a project, seeking feedback), its current state (such as current skill level, knowledge structure, and career stage), and the results of these actions (such as skill improvement, project success, and promotion opportunities, which are quantified by a reward function). It uses a random replay mechanism: Unlike chronological learning, the experience replay pool randomly draws past experience samples from it for learning. The agent is the subject of decision-making and action execution, as well as the subject of learning and adaptation. It can perceive the current state of professional development (state s), such as understanding its own skill shortcomings, market demand for a certain skill, and available learning resources. Based on the current state and the knowledge (strategy) learned from the experience replay pool, the agent selects an action (action a), such as which new skill to learn, which project to apply for, or who to seek guidance from.

[0106] The reward function defines what behaviors and results are valuable, provides goals and direction for the agent's learning, and converts various "good results" or "bad results" in professional development into a numerical value (reward r). For example, successfully completing a challenging project may earn +5 points, learning a new skill that is in high demand in the market may earn +3 points, missing a promotion opportunity may earn -2 points, and wasting time on irrelevant activities may earn -1 point. Learning is achieved by maximizing long-term cumulative rewards. The reward function acts like a "mentor," telling the agent which behaviors lead to positive outcomes and which behaviors to avoid. The design of the reward function directly determines the strategy learned by the agent. The goal definition clearly defines the goal pursued by the model (or individual). In the professional development scenario, this goal may be skill improvement, career advancement, job satisfaction, income growth, or some combination of these goals.

[0107] Based on the teacher professional development evaluation platform, teacher evaluation is conducted based on the collected real-time multimodal data of teachers to obtain real-time teacher scores and predicted teacher scores, including the following steps:

[0108] S1-1: Collect teachers' real-time multimodal data, pre-process the real-time multimodal data, and store the pre-processed real-time multimodal data in the teacher professional development evaluation platform;

[0109] Preprocessing includes data cleaning, missing value processing (interpolation method), outlier detection and feature selection (mutual information method) in sequence, and normalization of structured data;

[0110] Real-time multimodal data includes class hours and student grades from the teaching management system (structured data), teaching reflections and training records from teacher development files (semi-structured data), interactive behavior analysis from classroom videos (unstructured data), and education policy documents from industry reports (external data).

[0111] S1-2: Based on the teacher professional development evaluation platform, use the intelligent evaluation model to evaluate teachers based on the pre-processed real-time multimodal data to obtain real-time three-level indicator teacher scores, including the following steps:

[0112] S1-2-1: Based on the teacher professional development evaluation platform, use the multimodal data feature extraction module of the intelligent evaluation model to extract real-time multimodal data features of the pre-processed real-time multimodal data;

[0113] S1-2-2: Based on the preset attention weight value, the feature fusion module of the intelligent evaluation model is used to perform weighted fusion on the real-time multimodal data features to obtain real-time weighted fusion features;

[0114] S1-2-3: Based on the real-time weighted fusion features, the intelligent evaluation module of the intelligent evaluation model is used to evaluate teachers and obtain real-time three-level indicator teacher scores;

[0115] S1-3: Based on the real-time three-level indicator teacher ratings, a multi-dimensional evaluation indicator mathematical model is used to generate corresponding real-time teacher ratings, including the following steps:

[0116] S1-3-1: Input the real-time third-level indicator teacher rating into the second-level indicator aggregation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time second-level indicator teacher rating;

[0117] S1-3-2: Input the real-time second-level indicator teacher rating into the first-level indicator aggregation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time first-level indicator teacher rating;

[0118] S1-3-3: Input the real-time first-level indicator teacher rating into the comprehensive evaluation score mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time comprehensive teacher rating;

[0119] Input the real-time first-level indicator teacher rating and the real-time comprehensive teacher rating into the value-added evaluation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time development increment;

[0120] S1-3-4: Integrate the real-time first-level indicator teacher rating, the real-time comprehensive teacher rating, and the real-time development increment to obtain the corresponding real-time teacher rating;

[0121] S1-4: Based on the real-time teacher rating, use the rating prediction model to perform rating prediction to obtain the predicted teacher rating, including the following steps:

[0122] S1-4-1: Use the time series feature extraction module of the rating prediction model to extract the real-time time series features of the real-time first-level indicator teacher rating in the real-time teacher rating;

[0123] S1-4-2: Based on the real-time time series features, the score prediction module of the score prediction model is used to perform score prediction and obtain the predicted first-level indicator teacher score;

[0124] S1-4-3: Input the predicted first-level indicator teacher score into the comprehensive evaluation score mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the predicted comprehensive teacher score;

[0125] S1-4-4: Input the predicted first-level indicator teacher score and the predicted comprehensive teacher score into the value-added evaluation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the predicted development increment;

[0126] S1-4-5: Integrate the predicted first-level indicator teacher score, the predicted comprehensive teacher score, and the predicted development increment to obtain the corresponding predicted teacher score;

[0127] S2: Generate personalized development suggestions based on real-time teacher ratings and predicted teacher ratings, obtain real-time personalized development suggestions, and visualize them. This includes the following steps:

[0128] S2-1: If the real-time first-level indicator teacher score in the real-time teacher score is lower than the score threshold, proceed to the next step; otherwise, continue the teacher evaluation process;

[0129] S2-2: Input real-time teacher ratings and predicted teacher ratings into the professional development recommendation model of the teacher professional development evaluation platform;

[0130] S2-3: updating the state space of the agent of the professional development suggestion model according to the real-time first-level indicator teacher rating in the real-time teacher rating to obtain an updated state space;

[0131] The updated state space of the agent in the professional development recommendation model is defined as the teacher's current ability vector (the real-time first-level indicator teacher rating in the real-time teacher rating);

[0132] Action Space includes: training resources: teacher ethics training, teaching method innovation, data technology application and other categories; teaching and research activities: school-level open classes, regional teaching and research forums, cross-school collaborative projects, etc.

[0133] S2-4: updating the reward function of the professional development recommendation model according to the predicted development increment in the predicted teacher rating to obtain an updated reward function;

[0134] S2-5: Based on the updated action space and the updated reward function, the agent of the professional development suggestion model is used to generate personalized development suggestions and obtain real-time personalized development suggestions;

[0135] S2-6: Store and visualize teachers’ real-time teacher ratings, predicted teacher ratings, and real-time personalized development suggestions.

[0136] Example 2:

[0137] like Figure 2 As shown, this embodiment provides a teacher evaluation system based on multimodal data, which is used to implement a teacher evaluation method. The system includes a teacher professional development evaluation platform and an external terminal, and the teacher professional development evaluation platform is communicatively connected to the external terminal;

[0138] The teacher professional development evaluation platform is built based on a layered distributed architecture, which includes an infrastructure layer, a data middle platform layer, an application service layer, and a user interaction layer. The user interaction layer is connected to external terminals for communication.

[0139] Preferably, the infrastructure layer includes a distributed computing module and a storage module;

[0140] Computing resources of the distributed computing module: Kubernetes, a container orchestration system, is used to manage Docker container clusters, allocating central processing unit (CPU) and memory resources on demand and supporting elastic scaling (peak processing capacity reaches 100,000 teachers / hour).

[0141] Storage architecture of the storage module: Structured data storage architecture: Uses HBase to store basic teacher information, evaluation results, etc., supporting high-concurrency read and write (queries per second (QPS) > 5000); Unstructured data storage architecture: Uses the Hadoop Distributed File System (HDFS) to store classroom videos, documents, etc., with a single cluster storage capacity of terabytes (PB); Graph data storage architecture: Builds a teacher development knowledge graph based on Neo4j, supporting real-time relational queries (response time < 500ms);

[0142] The data middle platform layer includes the data governance module, knowledge graph module, and user portrait module;

[0143] The data governance module is used for connecting data sources, data cleaning, format unification / standardized interfaces, quality verification / rule engine, data desensitization / privacy removal, metadata management / data catalog;

[0144] The knowledge graph module is used to store knowledge graphs, which contain professional knowledge graph data on teachers and education. It is used to define and explain entity types (teachers, courses, achievements, training, etc.), relationship types (lectures, publications, participation, etc.) and professional names of attributes (such as teacher titles, course difficulty, achievement levels, etc.);

[0145] The user portrait module is used to construct the teacher's user portrait based on the portrait structure and corresponding professional name defined by the knowledge graph module. For example, Teacher Zhang → Lectures → Course: High School Chinese; Teacher Zhang → Publishes → Papers: Core Journals; Teacher Zhang → Participates → Training: Information-based Teaching;

[0146] The application service layer includes intelligent evaluation service module and decision support service module;

[0147] The intelligent evaluation service module is used to provide teacher evaluation services. It evaluates teachers based on the collected real-time multimodal data of teachers and obtains real-time teacher scores and predicted teacher scores.

[0148] The decision support service module is used to provide decision support services, generate personalized development suggestions based on real-time teacher scores and predicted teacher scores, and obtain real-time personalized development suggestions;

[0149] The user interaction layer includes the teacher client and the management client;

[0150] The teacher client provides functions such as visual display, growth file management, real-time evaluation query, and personalized suggestion reception. The interface adopts a card-style design;

[0151] The management client supports functions such as visual display, regional data statistics, weak link warning, policy simulation and deduction, and includes components such as heat maps and trend comparison charts.

[0152] The present invention provides a teacher evaluation method and system based on multimodal data. The multi-dimensional evaluation index mathematical model constructed provides a city teacher professional development index system and constructs a three-dimensional evaluation framework including several first-level indicators, several second-level indicators and several third-level indicators. It can break through the limitations of traditional single evaluation dimension, reflect the real teaching status and professional performance of teachers from a more comprehensive and multi-angle perspective, reduce subjective bias, and improve the objectivity and comprehensiveness of evaluation; based on educational measurement theory, it integrates advanced technologies such as deep learning, knowledge graphs, and federated learning to construct a three-dimensional support system of "scientific indicators, intelligent evaluation, and scenario-based application"; with "multimodal data" as the core input, and processed through professional models such as intelligent evaluation models and scoring prediction models, it effectively integrates and deeply mines various data generated in the classroom (including traditional methods The system can collect and analyze the unstructured data ignored by teachers, transform these data into valuable evaluation information and decision-making basis, and reflect the advantages of technology-driven development. Based on real-time scoring and predictive scoring, the professional development suggestion model generates "real-time personalized development suggestions". These suggestions can be closely combined with the current specific performance of teachers and future development trends, and provide more targeted and feasible improvement directions and development paths, effectively solving the problem of "one size fits all", and better meeting the professional growth needs of individual teachers. Based on the teacher professional development evaluation platform, it realizes automated and intelligent data collection, intelligent evaluation and personalized suggestions, reduces the workload of evaluators, and realizes multi-party linkage through teacher clients and management clients, building an evaluation ecosystem with data interoperability, functional collaboration, and covering teachers and managers, thereby improving the overall efficiency and coordination of evaluation work.

[0153] The present invention is not limited to the above optional embodiments. Anyone can derive various other forms of products based on the teachings of the present invention. The above specific embodiments should not be construed as limiting the scope of protection of the present invention. The scope of protection of the present invention shall be based on the scope defined in the claims, and the description can be used to interpret the claims.

Claims

1. A teacher evaluation method based on multimodal data, characterized by: The steps include: Based on the teacher professional development evaluation platform, teachers are evaluated based on the collected real-time multimodal data of teachers, and real-time teacher scores and predicted teacher scores are obtained; Based on real-time teacher ratings and predicted teacher ratings, personalized development suggestions are generated, real-time personalized development suggestions are obtained, and visually displayed.

2. The teacher evaluation method based on multimodal data according to claim 1, characterized in that: The teacher professional development evaluation platform is equipped with a multi-dimensional evaluation indicator mathematical model, an intelligent evaluation model, a scoring prediction model and a professional development suggestion model.

3. The teacher evaluation method based on multimodal data according to claim 2, characterized in that: The multi-dimensional evaluation index mathematical model includes a secondary index aggregation mathematical sub-model, a primary index aggregation mathematical sub-model, a comprehensive evaluation score mathematical sub-model and a value-added evaluation mathematical sub-model; The intelligent evaluation model includes a multimodal data feature extraction module, a multi-task learning module, a feature fusion module and an intelligent evaluation module; The rating prediction model includes a time series feature extraction module and a rating prediction module; The professional development suggestion model includes an experience replay pool, an intelligent agent, and a reward function.

4. The teacher evaluation method based on multimodal data according to claim 3, characterized in that: Based on the teacher professional development evaluation platform, teacher evaluation is conducted based on the collected real-time multimodal data of teachers to obtain real-time teacher scores and predicted teacher scores, including the following steps: Collect teachers' real-time multimodal data, pre-process the real-time multimodal data, and store the pre-processed real-time multimodal data in the teacher professional development evaluation platform; Based on the teacher professional development evaluation platform, an intelligent evaluation model is used to evaluate teachers based on pre-processed real-time multimodal data, obtaining real-time three-level indicator teacher scores; Based on the real-time three-level indicator teacher ratings, a multi-dimensional evaluation indicator mathematical model is used to generate corresponding real-time teacher ratings; Based on the real-time teacher ratings, the rating prediction model is used to make rating predictions and obtain the predicted teacher ratings.

5. The teacher evaluation method based on multimodal data according to claim 4, characterized in that: Based on the teacher professional development evaluation platform, an intelligent evaluation model is used to evaluate teachers based on pre-processed real-time multimodal data, obtaining real-time three-level indicator teacher scores, including the following steps: Based on the teacher professional development evaluation platform, the multimodal data feature extraction module of the intelligent evaluation model is used to extract the real-time multimodal data features of the pre-processed real-time multimodal data; According to the preset attention weight value, the feature fusion module of the intelligent evaluation model is used to perform weighted fusion on the real-time multimodal data features to obtain real-time weighted fusion features; According to the real-time weighted fusion features, the intelligent evaluation module of the intelligent evaluation model is used to evaluate teachers and obtain real-time three-level indicator teacher evaluations.

6. The teacher evaluation method based on multimodal data according to claim 5, characterized in that: Based on the real-time three-level indicator teacher ratings, a multi-dimensional evaluation indicator mathematical model is used to generate the corresponding real-time teacher ratings, including the following steps: Input the real-time third-level indicator teacher ratings into the second-level indicator aggregation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time second-level indicator teacher ratings; Input the real-time second-level indicator teacher rating into the first-level indicator aggregation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time first-level indicator teacher rating; Input the real-time first-level indicator teacher rating into the comprehensive evaluation score mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time comprehensive teacher rating; Input the real-time first-level indicator teacher rating and the real-time comprehensive teacher rating into the value-added evaluation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the real-time development increment; Integrate the real-time first-level indicator teacher rating, real-time comprehensive teacher rating and real-time development increment to obtain the corresponding real-time teacher rating.

7. The teacher evaluation method based on multimodal data according to claim 6, characterized in that: Based on the real-time teacher ratings, the rating prediction model is used to predict the ratings and obtain the predicted teacher ratings, including the following steps: Use the time series feature extraction module of the rating prediction model to extract the real-time time series features of the real-time first-level indicator teacher rating in the real-time teacher rating; According to the real-time time series characteristics, the score prediction module of the score prediction model is used to perform score prediction and obtain the predicted first-level indicator teacher score; Input the predicted first-level indicator teacher score into the comprehensive evaluation score mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the predicted comprehensive teacher score; Input the predicted first-level indicator teacher score and the predicted comprehensive teacher score into the value-added evaluation mathematical sub-model of the multi-dimensional evaluation indicator mathematical model to obtain the predicted development increment; Integrate the predicted first-level indicator teacher rating, the predicted comprehensive teacher rating, and the predicted development increment to obtain the corresponding predicted teacher rating.

8. The teacher evaluation method based on multimodal data according to claim 7, characterized in that: Generate personalized development suggestions based on real-time teacher ratings and predicted teacher ratings, obtain real-time personalized development suggestions, and visualize them. This includes the following steps: If the real-time first-level indicator teacher score in the real-time teacher score is lower than the score threshold, then proceed to the next step; otherwise, continue with the teacher evaluation process; Input real-time teacher ratings and predicted teacher ratings into the professional development recommendation model of the teacher professional development evaluation platform; According to the real-time first-level indicator teacher rating in the real-time teacher rating, the state space of the intelligent agent of the professional development suggestion model is updated to obtain an updated state space; updating the reward function of the professional development recommendation model according to the predicted development increment in the predicted teacher ratings to obtain an updated reward function; Based on the updated action space and the updated reward function, the intelligent agent of the professional development suggestion model is used to generate personalized development suggestions and obtain real-time personalized development suggestions; Store and visualize teachers' real-time teacher ratings, predicted teacher ratings, and real-time personalized development suggestions.

9. A teacher evaluation system based on multimodal data, for implementing the teacher evaluation method according to any one of claims 1 to 8, characterized in that: The system includes a teacher professional development evaluation platform and an external terminal, wherein the teacher professional development evaluation platform is in communication connection with the external terminal; The teacher professional development evaluation platform is built based on a layered distributed architecture, which includes an infrastructure layer, a data middle platform layer, an application service layer and a user interaction layer. The user interaction layer is connected to the external terminal communication.

10. The teacher evaluation system based on multimodal data according to claim 9, characterized in that: The infrastructure layer includes a distributed computing module and a storage module; The data middle platform layer includes a data governance module, a knowledge graph module, and a user portrait module; The application service layer includes an intelligent evaluation service module and a decision support service module; The user interaction layer includes a teacher client and a management client.

Images