Dynamic evaluation method and system for teacher interdisciplinary literacy based on multi-dimensional data

Abstract

The application discloses a kind of based on multidimensional data's teacher interdisciplinary accomplishment dynamic evaluation method and system, wherein method includes: based on teacher subject background, teaching age, teaching period of study constructs evaluation standard library, and different accomplishment evaluation weight is assigned to each dimension;Build multidimensional data acquisition terminal, interface existing teaching research system, input teacher basic information and automatically match exclusive evaluation standard and weight;Real-time collection and multidimensional data of the interdisciplinary accomplishment of the teacher to be evaluated;The multidimensional data collected are preprocessed, and time series feature dataset is constructed;Based on the time series feature data, dynamic evaluation and trend prediction of teacher accomplishment are carried out, and the evaluation result of the interdisciplinary accomplishment of teacher is output in combination with the weight matched and current window dynamic threshold value.The significant effect is: evaluation accuracy is greatly improved, adapts to different scenes and teacher type, realizes the deep fusion of evaluation and improvement, and can help teachers to rapidly improve interdisciplinary accomplishment.

Description

Technical Field

[0001] This invention relates to the field of educational assessment technology, specifically to a method and system for dynamic assessment of teachers' interdisciplinary competence based on multi-dimensional data. Background Technology

[0002] With the deepening of curriculum reform in basic education, cultivating students' ability to solve complex problems across disciplines has become a core objective, which places higher demands on teachers' interdisciplinary competence. However, scientifically and effectively assessing teachers' interdisciplinary competence faces many challenges: First, existing assessment methods rely heavily on static data such as classroom observation, lesson plan review, and subjective questionnaires, lacking the collection of dynamic data such as teaching process, student feedback, and knowledge application, and thus failing to fully reflect the level of teacher competence. Secondly, existing assessment methods mostly use traditional methods such as manual statistics and weighted summation, which lack the collection of real, continuous, and multimodal data on the teaching process. Furthermore, they fail to achieve intelligent integration and in-depth mining of multi-source data, making them susceptible to subjective factors, resulting in insufficient evaluation evidence and questionable reliability and validity. Third, existing assessment methods are mostly periodic and one-off assessments, which cannot track the growth trajectory and development process of teachers' professional competence, make it difficult to provide targeted suggestions for improving their abilities, and make it difficult to evaluate the long-term effects of interventions such as training. Fourth, existing assessment methods only output assessment scores, which cannot reveal the specific causes of shortcomings in competency and cannot provide teachers with timely and accurate professional development guidance.

[0003] In view of this, there is an urgent need for a new method and system for assessing teachers' interdisciplinary competence that can integrate multi-dimensional process data, achieve dynamic tracking, and provide diagnostic feedback. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a dynamic assessment method and system for teachers' interdisciplinary competence based on multi-dimensional data. Through steps such as intelligent collection of multi-source data, analysis using a CNN-BiLSTM fusion model, and dynamic threshold adaptation, this method can achieve comprehensive, accurate, and real-time assessment of teachers' interdisciplinary competence, while also providing personalized improvement paths. This addresses the problems of subjectivity, staticity, and limited data in existing assessments, and provides technical support for improving the quality of interdisciplinary education.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: Firstly, this invention provides a dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data, the key of which includes the following steps: Step 1: Construct an assessment standard library based on teachers' subject background, teaching experience, and teaching stage, and assign different competency assessment weights to each dimension; Step 2: Set up a multi-dimensional data collection terminal, connect it to the existing teaching and research system, enter the teacher's basic information and automatically match exclusive evaluation standards and weights; Step 3: Real-time collection of multi-dimensional data on the interdisciplinary competence of teachers to be evaluated; Step 4: Perform data preprocessing on the collected multi-dimensional data to construct a time-series feature dataset; Step 5: Using the CNN-BiLSTM competency assessment model, dynamic assessment and trend prediction of teachers' competency are performed based on the time-series feature data, and the assessment results of teachers' interdisciplinary competency are output by combining the matched weights and the current window dynamic threshold. The CNN-BiLSTM literacy assessment model includes: The input layer is used to arrange the input time-series feature dataset in order of time windows; Feature enhancement layer, used to enhance the spatial features of the input data using a CNN network; The temporal analysis layer is used to perform bidirectional temporal analysis on the feature vectors output by the feature enhancement layer using a BiLSTM network to obtain the changing trend of teachers' interdisciplinary literacy over time. The output layer is used to generate assessment results of teachers' interdisciplinary competence based on the trend of teachers' interdisciplinary competence over time as output by the time series analysis layer.

[0006] Furthermore, the multi-dimensional data acquisition terminal includes intelligent recording and broadcasting equipment, online question-answering equipment, student feedback equipment, and teaching behavior analysis equipment.

[0007] Furthermore, the multi-dimensional data includes teaching process data, knowledge and ability data, student feedback data, teaching and research achievement data, process behavior data, and external evaluation data, wherein: The teaching process data includes classroom video data, classroom audio data, and text data such as courseware, lesson plans, and blackboard writing. The knowledge and ability data includes teachers' knowledge reserve data and knowledge integration ability data; The student feedback data includes data on classroom satisfaction, knowledge mastery, and acceptance of interdisciplinary teaching.

[0008] Furthermore, step 4 involves preprocessing the collected multi-dimensional data to construct a time-series feature dataset, which includes the following steps: Step 4.1: Use the 3σ criterion to remove outlier data and use a missing value imputation algorithm to fill in the missing data; Step 4.2: Normalize and standardize the data from different dimensions; Step 4.3: Use the random forest algorithm to select core features that are strongly related to interdisciplinary literacy, and then fuse the selected multimodal features through feature concatenation to form an assessment dataset; Step 4.4: Divide the data in the assessment dataset into segments according to the time dimension, and construct a time-series feature dataset corresponding to each time window.

[0009] Furthermore, the normalization standardization process for data of different dimensions described in step 4.2 specifically includes: Map numerical data to the [0,1] interval; Feature extraction from text data; Audio data feature extraction based on Mel frequency cepstral coefficients; Feature extraction of video data is performed based on the YOLOv8-nano model.

[0010] Furthermore, the process of feature extraction from video data based on the YOLOv8-nano model is as follows: The classroom videos are downsampled at a preset frequency and simultaneously scaled to a preset resolution. The YOLOv8-nano model is used to detect and extract key action features of teachers and students in the classroom, and output action categories, bounding box coordinates and confidence scores. Global feature encoding is performed on the detected video frames, and video feature vectors are generated by combining action timing information.

[0011] Furthermore, the feature enhancement layer uses a 4-layer CNN network to enhance spatial features layer by layer; the temporal analysis layer uses a two-layer BiLSTM network to perform bidirectional temporal analysis on the feature vectors output by the feature enhancement layer; the output layer uses two fully connected layers to reduce the dimensionality of the temporal features and perform nonlinear transformation, and uses the Softmax activation function to convert the output results into a probability distribution.

[0012] Furthermore, the dynamic threshold is dynamically updated using a sliding window algorithm, including the following steps: Set the sliding window length, divide the window period, each window corresponds to a set of regional teacher competency datasets of the same type, obtain the benchmark dataset, calculate the initial threshold, and set the threshold range; Retrieve valid evaluation data of teachers of the same type in the region within a certain period, and combine it with the historical valid data from the previous window to form the statistical data set for the current window; Based on the statistical data set, calculate the mean and standard deviation of the overall competence scores of teachers of the same type, and calculate a new threshold based on the mean and standard deviation; Dynamic threshold updates are achieved by replacing the initial or old threshold with a new one.

[0013] Furthermore, the evaluation results include the evaluation level, detailed scores for each dimension, strengths and weaknesses, and data source labeling.

[0014] Secondly, the present invention provides a dynamic assessment system for teachers' interdisciplinary competence based on multi-dimensional data for implementing the method described in the first aspect, the key feature of which is: The standard establishment and weight allocation module is used to build an assessment standard library based on teachers' subject background, teaching experience, and teaching stage, and to assign different competency assessment weights to each dimension; The data platform building module is used to build a multi-dimensional data collection terminal and connect to the existing teaching and research system to input teachers' basic information and automatically match exclusive evaluation standards and weights. The data acquisition module is used to collect multi-dimensional data on the interdisciplinary competence of teachers to be evaluated in real time. The data preprocessing module is used to preprocess the collected multi-dimensional data and construct a time-series feature dataset; The dynamic assessment module is used to perform dynamic assessment and trend prediction of teachers' competence based on the time-series feature data using the CNN-BiLSTM competence assessment model, and output the assessment results of teachers' interdisciplinary competence by combining the matched weights and the current window dynamic threshold.

[0015] The significant effects of this invention are: 1. This invention collects six categories of multi-dimensional data covering the core dimensions of teachers' interdisciplinary literacy, and combines them with a CNN-BiLSTM fusion model to deeply mine features, effectively avoiding the influence of subjective factors and significantly improving the accuracy of the assessment compared to existing technologies.

[0016] 2. This invention, by constructing a time-series dataset and a dynamic threshold system, tracks the changing trends of teachers' competence in real time, and matches exclusive assessment standards with individual differences among teachers, thus realizing dynamic and personalized assessment. It effectively solves the limitations of traditional static and generalized assessment and can be adapted to different scenarios and teacher types.

[0017] 3. This invention can provide school administrators with data support such as an overall analysis of regional teacher competence and the distribution of shortcomings, helping to accurately carry out teaching and research activities, optimize teacher allocation, and promote the effective implementation of interdisciplinary education reforms. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method described in this invention; Figure 2 This is a flowchart of data preprocessing; Figure 3 This is a schematic diagram of the CNN-BiLSTM literacy assessment model; Figure 4This is a schematic diagram of the system described in this invention. Detailed Implementation

[0019] The specific embodiments and working principles of the present invention will be further described in detail below with reference to the accompanying drawings.

[0020] Example 1: like Figure 1 As shown in the figure, this embodiment proposes a dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data. The specific steps are as follows: Step 1: Construct an assessment standard library based on teachers' subject background, teaching experience, and teaching stage, and assign different competency assessment weights to each dimension; In implementation, the teacher's subject background can be humanities, sciences, or arts; teaching experience can be categorized as new teacher, young teacher, or senior teacher; and teaching level can be primary school, middle school, or university. A personalized assessment standard library is constructed based on these three dimensions: teacher's subject background, teaching experience, and teaching level. When allocating weights, the following can be followed: 30% weight for knowledge integration for science teachers, and 30% weight for classroom interaction for humanities teachers.

[0021] Step 2: Set up a multi-dimensional data collection terminal, connect it to the existing teaching and research system, enter the teacher's basic information and automatically match exclusive evaluation standards and weights; In this example, the multi-dimensional data acquisition terminal includes intelligent recording and broadcasting equipment, online quiz equipment, student feedback equipment, and teaching behavior analysis equipment. It collects data using both online and offline parallel methods, including 4K classroom video, 48kHz audio, text courseware, knowledge quiz data, student feedback, teaching and research results, and teaching behavior data. Thus, by using the multi-dimensional data acquisition terminal and a software platform that connects to existing teaching and research systems, integrated hardware and software data acquisition is achieved. Combined with offline acquisition methods, this enables full-scenario, interference-free acquisition of assessment data, adaptable to both online and offline teaching scenarios.

[0022] In specific implementation, this embodiment divides the collected data into six categories: teaching process data, knowledge and ability data, student feedback data, teaching and research results data, process behavior data, and external evaluation data, as detailed below: The teaching process data includes classroom video data, classroom audio data, and text data such as courseware, lesson plans, and blackboard writing, which are collected through intelligent recording and broadcasting terminals equipped with 4K high-definition cameras and omnidirectional microphones, as well as teaching terminals such as tablets or computers. The knowledge and ability data includes teachers' knowledge reserve data and knowledge integration ability data; the collection method is: to build an interdisciplinary knowledge question bank covering multiple disciplines such as literature, science, art, and engineering, and collect the data through an online answering terminal; The student feedback data includes data on classroom satisfaction, knowledge mastery, and recognition of interdisciplinary teaching. This type of data is collected through student terminals such as mobile phones, tablets, and computers. When collecting the data, a combination of star rating and text evaluation can be used to ensure the authenticity of the feedback.

[0023] The teaching and research achievements data include teachers' interdisciplinary teaching and research papers, research projects, teaching case awards, and other achievements data, which are extracted quantitatively according to the achievement level and influence, and are obtained by connecting to the school's teaching and research management system during collection; The process behavior data includes behavioral data such as the number of times teachers ask questions in class, the number of times they extend interdisciplinary knowledge points, and the duration of group guidance, as well as the duration and frequency of teachers' participation in interdisciplinary training and teaching research activities. The data is collected through a teaching behavior analysis terminal. The external evaluation data includes expert comments on teachers' interdisciplinary instructional design, classroom implementation, and outcome transformation, which are collected by connecting to an expert review system.

[0024] Step 3: Real-time collection of multi-dimensional data on the interdisciplinary competence of teachers to be evaluated; In practice, the frequency of data collection for various types is suppressed, and the data is encrypted before transmission to ensure data security.

[0025] Step 4: Perform data preprocessing on the collected multi-dimensional data to construct a time-series feature dataset; In the specific implementation process, the process of preprocessing the collected multi-dimensional data to construct a time-series feature dataset is as follows: Figure 2 As shown, it includes the following steps: Step 4.1: Use the 3σ criterion to remove outlier data and use a missing value imputation algorithm to fill in the missing data; In practice, numerical data is filled with the mean, and text data is filled with the mode. Step 4.2: Normalize and standardize the data from different dimensions; Since the collected data includes different types of data such as text, audio, and video, the specific steps for normalizing and standardizing data of different dimensions are as follows: Numerical data such as the accuracy of answering questions and the number of interactions are mapped to the [0,1] interval; For text, audio, and video—three types of non-numerical data—features are extracted using dedicated algorithms. Specifically: After word segmentation and stop word removal, feature extraction is performed using the Word2Vec model. The specific process is as follows: First, the text data, such as lesson plans and comments, is segmented into words, and stop words, punctuation marks, and special characters are removed, while retaining core semantic vocabulary. Then, a Word2Vec model is constructed using the Skip-gram pattern, with a window size of 5, a vector dimension of 128, 50 iterations, and a minimum word frequency threshold of 3, converting the preprocessed text vocabulary into 128-dimensional word vectors. Finally, the average of all word vectors in a single text is taken to obtain a 128-dimensional text feature vector, achieving a quantitative representation of the text semantics. After pre-emphasis, frame segmentation, and windowing, the audio data is used to extract features based on Mel-frequency cepstral coefficients. The specific process is as follows: First, the audio data is pre-emphasized to filter low-frequency noise. Then, the audio is segmented into a frame sequence with a frame length of 25ms and a frame shift of 10ms, and a Hanning window is superimposed on each frame to suppress spectral leakage. Finally, the logarithm of the Mel spectrum is taken and a discrete cosine transform is performed to extract the first 13 MFCC coefficients. At the same time, the first and second order difference coefficients are calculated to form a 39-dimensional audio feature vector, which captures the intonation, speech rate and other features of teacher-student interactive speech. Feature extraction from video data is performed based on the YOLOv8-nano model. The specific process is as follows: First, the video data is downsampled at a frequency of 1 frame every 5 frames to reduce computational load, while being uniformly scaled to a resolution of 640×640. Then, the YOLOv8-nano model is used to detect key actions of teachers and students in the classroom, outputting action categories, bounding box coordinates, and confidence scores. Finally, global feature encoding is performed on the detected video frames, and a 256-dimensional video feature vector is generated by combining action temporal information to accurately capture interactive scenes and teaching behaviors in interdisciplinary classrooms.

[0026] Step 4.3: Use the random forest algorithm to screen core features that are strongly related to interdisciplinary literacy, remove redundant features, set the screening threshold to 0.7, and fuse the screened multimodal features (numerical type 128 dimensions, text type 128 dimensions, audio type 39 dimensions, video type 256 dimensions, totaling 551 dimensions) through feature concatenation, and then map them through a fully connected layer to form a 1024-dimensional unified feature vector to form the assessment dataset. Step 4.4: Divide the data in the evaluation dataset into segments according to the time dimensions of day, week, and month. Each time window corresponds to a set of feature vectors to construct a time-series feature dataset.

[0027] Step 5: Using the CNN-BiLSTM competency assessment model, dynamic assessment and trend prediction of teachers' competency are performed based on the time-series feature data, and the assessment results of teachers' interdisciplinary competency are output by combining the matched weights and the current window dynamic threshold. In this example, the structure of the CNN-BiLSTM literacy assessment model is as follows: Figure 3 As shown, it specifically includes: The input layer is used to arrange the 1024-dimensional multimodal temporal feature vectors generated by the data preprocessing layer into input data in the format of [batch_size, time_step, feature_dim] according to the time window order. The batch_size is set to 32, the time_step is set to 7, corresponding to one set of feature vectors per week and 7 consecutive weeks of data, and feature_dim=1024. The feature enhancement layer is used to enhance spatial features layer by layer using a 4-layer CNN network. The first CNN layer uses 64 3×3 convolutional kernels with a stride of 1 to perform convolution operations on the spatial dimension of the input temporal features. After ReLU activation, it is connected to a BatchNorm layer to suppress overfitting. The second CNN layer uses 128 3×3 convolutional kernels and repeats the convolution, ReLU activation, and BatchNorm process to further extract deep spatial correlation features. The third and fourth layers each use 256 2×2 convolutional kernels with a stride of 1. After feature enhancement, the feature map is downsampled through a 2×2 max pooling layer. Finally, the 1024-dimensional input features are compressed into a 256-dimensional feature vector, preserving the core features while reducing the computational cost. The temporal analysis layer is used to input the 256-dimensional feature vector output by the feature enhancement layer into a two-layer BiLSTM network. The first layer of the BiLSTM network has 128 neurons and a dropout coefficient of 0.2. The forward LSTM captures the feature dependencies from the past to the present in the time series, and the backward LSTM captures the feature associations from the present to the future, outputting a 256-dimensional bidirectional feature fusion result. The second layer of the BiLSTM network also uses 128 neurons and a dropout coefficient of 0.2 to further mine the temporal patterns of the bidirectional features, outputting a 128-dimensional temporal feature vector, accurately capturing the changing trend of teachers' interdisciplinary literacy over time. The output layer is used to generate assessment results of teachers' interdisciplinary competence based on the trend of teachers' interdisciplinary competence over time as output by the time series analysis layer.

[0028] In this example, the output layer is implemented using two fully connected layers. The first fully connected layer has 64 neurons and uses the ReLU activation function to perform dimensionality reduction and nonlinear transformation of temporal features. The second fully connected layer has 4 neurons and uses the Softmax activation function to convert the output result into a probability distribution. The category with the highest probability is the teacher competence level. At the same time, the score of each competence dimension is output. This score is calculated based on the weighted average of the intermediate outputs of the fully connected layer and ranges from 0 to 100.

[0029] Based on the above model structure, the training steps of the CNN-BiLSTM literacy assessment model are as follows: First, we organized 10,000+ sets of historical assessment data, covering multiple disciplines including humanities, sciences, arts, and engineering, with teaching experience ranging from 1 to 30 years, and covering all educational stages from primary school to university. The data was divided into pre-training set, pre-validation set, and pre-test set in a 7:2:1 ratio, and each set of data was processed into [32,7,1024] grids. Secondly, the model parameters were initialized using the He normal initialization strategy, the optimizer was AdamW, the initial learning rate was 0.001, and the loss function was the cross-entropy loss function. The number of iterations was set to 500, the batch size to 32, and the pre-training set loss and pre-validation set accuracy were calculated after each iteration. The learning rate was dynamically adjusted using the cosine annealing strategy. An early stopping mechanism was added: if the pre-validation set accuracy did not improve for 20 consecutive iterations (fluctuation ≤ 0.5%), pre-training was stopped immediately, and the optimal pre-trained model parameters were saved. Finally, based on the pre-trained model, real-time data collected from teachers in the current region was imported, the parameters of the first two CNN layers of the feature enhancement layer were frozen, and only the parameters of the last two CNN layers, BiLSTM, and fully connected layers were fine-tuned. During the fine-tuning stage, the learning rate was reduced to 5e-4, the number of iterations was 200, and the batch size was 32. The accuracy of the validation set of real-time data was used as the core indicator to ensure that the accuracy of the fine-tuned model in the target scenario was ≥95%, and the model initialization and adaptation were completed.

[0030] The inference process of the trained CNN-BiLSTM literacy assessment model is as follows: First, retrieve the teacher time-series feature dataset for the past 7 days from the cloud server. If the data volume is less than 7 days, use the forward padding method to complete it to ensure the integrity of the input data. Then, the current optimal model parameters are loaded, and forward propagation calculation is performed on the input data. After feature enhancement, time series analysis, and Softmax transformation, the competency level and scores of each dimension of each teacher are output. At the same time, the confidence level of the inference result is calculated. If the confidence level is <0.8, it is automatically marked as a suspicious result and the manual review process is triggered. Afterwards, newly collected assessment data is automatically collected every 24 hours. Valid data is filtered and added to the training set. After training in incremental training mode, the assessment accuracy of the new model and the original model is calculated. If the accuracy of the new model increases by ≥1%, the model parameters are updated and overwritten. Otherwise, the original parameters are retained to ensure that the model always adapts to changes in teacher competence and maintains stable assessment accuracy. Finally, after each parameter update, historical model parameters are automatically backed up. If an evaluation anomaly occurs, the model can be reverted to its historical best, and a model fault investigation process can be triggered to locate data or parameter problems.

[0031] Step 6: Generate a personalized assessment report and improvement plan based on the assessment results, push them to relevant personnel, and track the improvement progress; and update model parameters, dynamic thresholds, and improvement plans based on the improvement data and retest results to achieve continuous optimization.

[0032] When generating the evaluation report, the scores of each dimension are output based on the inference of the CNN-BiLSTM model. The overall score is calculated by weighting according to the personalized weights. The literacy level is determined by comparing with the dynamic threshold. Then, an evaluation report is generated that includes the overall level, detailed scores of each dimension, strengths and weaknesses (dimensions with scores below the average are judged as weaknesses), and data source labels.

[0033] Therefore, to form a closed-loop improvement system and overcome the bottleneck of existing technologies that only assess but do not improve, this embodiment automatically generates customized improvement suggestions based on the assessment results and the teacher's skill gaps. These suggestions include recommendations for interdisciplinary training courses, matching of teaching and research activities, and reference to teaching cases. For example, for teachers with insufficient knowledge integration capabilities, interdisciplinary knowledge integration training courses are recommended; for teachers with insufficient classroom interaction, interdisciplinary classroom interaction design cases are provided. Furthermore, assessment reports and improvement plans are pushed to teachers and school administrators through a human-computer interaction terminal, allowing teachers to view skill change curves and progress on improvement tasks. Administrators can view the overall skill level and skill gap distribution of teachers in the region, providing a basis for teaching and research management decisions. A comprehensive reassessment is conducted every three months to compare the assessment results with the improvement goals and dynamically adjust the improvement plan. Data from the improvement process is fed back to a multi-dimensional data collection terminal to enrich the assessment dataset, optimize model parameters, and form a closed-loop iteration. Thus, it not only tracks improvement progress based on automatically generated personalized improvement plans but also continuously adapts to changes in teacher capabilities through iterative optimization, achieving a deep integration of assessment and improvement, effectively helping teachers rapidly improve their interdisciplinary skills.

[0034] In its specific implementation, this embodiment employs a quadratic polynomial fitting algorithm during the teaching competence improvement process. Based on the overall teacher competence scores and temporal characteristics changes over the past three windows, a trend fitting curve is constructed to predict the competence change trend for the next three windows. If the predicted score shows a continuous downward trend, a competence decline warning is triggered, the warning level is marked in the report, and the warning information is synchronized to the teacher's human-computer interaction terminal and the school administrator's backend. Generally, warnings are only pushed to the teacher, while severe warnings are also pushed to the administrator to remind them to intervene and provide guidance. In addition, when the assessment results show a sudden increase or decrease (fluctuation range ≥20%), the original data of the corresponding time window is automatically retrieved to locate abnormal reasons such as bonus points for teaching and research achievements or deduction points for teaching accidents, and these are marked in the assessment report to improve the transparency of the assessment.

[0035] In this embodiment, the dynamic threshold is dynamically updated using a sliding window algorithm, including the following steps: The sliding window length is set to 30 days, and the window period is divided into natural months. Each window corresponds to a set of regional teacher competency datasets of the same type. Teachers of the same type are defined as a group that is consistent with the target teacher in terms of subject, teaching experience, and grade level. The baseline dataset is obtained, the initial threshold is calculated, and the threshold range is set. After each 30-day window period ends, the system automatically retrieves valid evaluation data of teachers of the same type in the region within that period from the cloud server, and combines it with the historical valid data from the previous window to form the statistical data set for the current window. Based on the statistical data set, the overall competence score (μ) and standard deviation (σ) of teachers of the same type are calculated. Threshold ranges are set according to four competence levels: excellent level threshold is μ+1.2σ and above, good level threshold is μ+0.3σ to μ+1.2σ, ​​qualified level threshold is μ-0.3σ to μ+0.3σ, and level that needs improvement threshold is μ-0.3σ and below. If the proportion of teachers of a certain level exceeds the preset range, the σ coefficient is finely adjusted to ensure a balanced distribution of each level and avoid threshold deviation. The new threshold replaces the initial or old threshold and serves as the basis for determining the competency level of all teachers of the same type in the next window. At the same time, the threshold adjustment log is recorded to achieve dynamic threshold updates.

[0036] Example 2: See appendix Figure 4 This invention proposes a dynamic assessment system for teachers' interdisciplinary competence based on multi-dimensional data to implement the method described in Embodiment 1, comprising: The standard establishment and weight allocation module is used to build an assessment standard library based on teachers' subject background, teaching experience, and teaching stage, and to assign different competency assessment weights to each dimension; The data platform building module is used to build a multi-dimensional data collection terminal and connect to the existing teaching and research system to input teachers' basic information and automatically match exclusive evaluation standards and weights. The data acquisition module is used to collect multi-dimensional data on the interdisciplinary competence of teachers to be evaluated in real time. The data preprocessing module is used to preprocess the collected multi-dimensional data and construct a time-series feature dataset; The dynamic assessment module is used to perform dynamic assessment and trend prediction of teachers' competence based on the time-series feature data using the CNN-BiLSTM competence assessment model, and output the assessment results of teachers' interdisciplinary competence by combining the matched weights and the current window dynamic threshold. The feedback improvement and closed-loop iteration module is used to generate personalized assessment reports and improvement plans based on the assessment results, push them to relevant personnel, and track the improvement progress; and to update model parameters, dynamic thresholds and improvement plans based on improvement data and retest results to achieve continuous optimization.

[0037] In summary, this invention collects multi-dimensional data covering six major categories of core dimensions of teachers' interdisciplinary competence. After data preprocessing, it utilizes a CNN-BiLSTM fusion model for deep feature mining and combines this with personalized assessment standards tailored to individual teacher differences. This enables dynamic and personalized assessment of teachers' interdisciplinary competence, generating personalized improvement plans, tracking improvement progress in real time, and updating model parameters, dynamic thresholds, and improvement plans based on improvement data and retest results for continuous optimization. Therefore, this invention not only effectively avoids the influence of subjective factors, significantly improving assessment accuracy compared to existing technologies, but also effectively overcomes the limitations of traditional static and generalized assessments. It is adaptable to different scenarios and teacher types, breaking through the technical bottleneck of traditional technologies that only assess but do not improve, achieving a deep integration of assessment and improvement. This can help teachers rapidly improve their interdisciplinary competence and provide school administrators with data support such as overall regional teacher competence analysis and weakness distribution, facilitating precise teaching and research activities, optimizing teacher allocation, and promoting the effective implementation of interdisciplinary education reform.

[0038] The technical solution provided by this invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make several improvements and modifications to this invention without departing from the principles of this invention, and these improvements and modifications also fall within the protection scope of the claims of this invention.

Claims

1. A dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data, characterized in that, Includes the following steps: Step 1: Construct an assessment standard library based on teachers' subject background, teaching experience, and teaching stage, and assign different competency assessment weights to each dimension; Step 2: Set up a multi-dimensional data collection terminal, connect it to the existing teaching and research system, enter the teacher's basic information and automatically match exclusive evaluation standards and weights; Step 3: Real-time collection of multi-dimensional data on the interdisciplinary competence of teachers to be evaluated; Step 4: Perform data preprocessing on the collected multi-dimensional data to construct a time-series feature dataset; Step 5: Using the CNN-BiLSTM competency assessment model, dynamic assessment and trend prediction of teachers' competency are performed based on the time-series feature data, and the assessment results of teachers' interdisciplinary competency are output by combining the matched weights and the current window dynamic threshold. The CNN-BiLSTM literacy assessment model includes: The input layer is used to arrange the input time-series feature dataset in order of time windows; Feature enhancement layer, used to enhance the spatial features of the input data using a CNN network; The temporal analysis layer is used to perform bidirectional temporal analysis on the feature vectors output by the feature enhancement layer using a BiLSTM network to obtain the changing trend of teachers' interdisciplinary literacy over time. The output layer is used to generate assessment results of teachers' interdisciplinary competence based on the trend of teachers' interdisciplinary competence over time as output by the time series analysis layer.

2. The dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data according to claim 1, characterized in that: The multi-dimensional data acquisition terminal includes intelligent recording and broadcasting equipment, online question-answering equipment, student feedback equipment, and teaching behavior analysis equipment.

3. The dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data according to claim 2, characterized in that: The multi-dimensional data includes teaching process data, knowledge and ability data, student feedback data, teaching and research results data, process behavior data, and external evaluation data, among which: The teaching process data includes classroom video data, classroom audio data, and text data such as courseware, lesson plans, and blackboard writing. The knowledge and ability data includes teachers' knowledge reserve data and knowledge integration ability data; The student feedback data includes data on classroom satisfaction, knowledge mastery, and acceptance of interdisciplinary teaching.

4. The dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data according to claim 1, characterized in that: Step 4, which involves preprocessing the collected multi-dimensional data to construct a time-series feature dataset, includes the following steps: Step 4.1: Use the 3σ criterion to remove outlier data and use a missing value imputation algorithm to fill in the missing data; Step 4.2: Normalize and standardize the data from different dimensions; Step 4.3: Use the random forest algorithm to select core features that are strongly related to interdisciplinary literacy, and then fuse the selected multimodal features through feature concatenation to form an assessment dataset; Step 4.4: Divide the data in the assessment dataset into segments according to the time dimension, and construct a time-series feature dataset corresponding to each time window.

5. The dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data according to claim 4, characterized in that: Step 4.2, which describes the normalization and standardization process for data of different dimensions, specifically includes: Map numerical data to the [0,1] interval; Feature extraction from text data; Audio data feature extraction based on Mel frequency cepstral coefficients; Feature extraction of video data is performed based on the YOLOv8-nano model.

6. The dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data according to claim 5, characterized in that: The process of feature extraction from video data based on the YOLOv8-nano model is as follows: The classroom videos are downsampled at a preset frequency and simultaneously scaled to a preset resolution. The YOLOv8-nano model is used to detect and extract key action features of teachers and students in the classroom, and output action categories, bounding box coordinates and confidence scores. Global feature encoding is performed on the detected video frames, and video feature vectors are generated by combining action timing information.

7. The dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data according to claim 1, characterized in that: The feature enhancement layer uses a 4-layer CNN network to enhance spatial features layer by layer; The temporal analysis layer uses a two-layer BiLSTM network to perform bidirectional temporal analysis on the feature vectors output by the feature enhancement layer; the output layer uses two fully connected layers to perform dimensionality reduction and nonlinear transformation of the temporal features, and uses the Softmax activation function to convert the output results into a probability distribution.

8. The dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data according to claim 1, characterized in that: The dynamic threshold is dynamically updated using a sliding window algorithm, including the following steps: Set the sliding window length, divide the window period, each window corresponds to a set of regional teacher competency datasets of the same type, obtain the benchmark dataset, calculate the initial threshold, and set the threshold range; Retrieve valid evaluation data of teachers of the same type in the region within a certain period, and combine it with the historical valid data from the previous window to form the statistical data set for the current window; Based on the statistical data set, calculate the mean and standard deviation of the overall competence scores of teachers of the same type, and calculate a new threshold based on the mean and standard deviation; Dynamic threshold updates are achieved by replacing the initial or old threshold with a new one.

9. The dynamic assessment method for teachers' interdisciplinary competence based on multi-dimensional data according to claim 1, characterized in that: The evaluation results include the evaluation level, detailed scores for each dimension, strengths and weaknesses, and data source labeling.

10. A dynamic assessment system for teachers' interdisciplinary competence based on multi-dimensional data, used to implement the method described in any one of claims 1-9, characterized in that, include: The standard establishment and weight allocation module is used to build an assessment standard library based on teachers' subject background, teaching experience, and teaching stage, and to assign different competency assessment weights to each dimension; The data platform building module is used to build a multi-dimensional data collection terminal and connect to the existing teaching and research system to input teachers' basic information and automatically match exclusive evaluation standards and weights. The data acquisition module is used to collect multi-dimensional data on the interdisciplinary competence of teachers to be evaluated in real time. The data preprocessing module is used to preprocess the collected multi-dimensional data and construct a time-series feature dataset; The dynamic assessment module is used to perform dynamic assessment and trend prediction of teachers' competence based on the time-series feature data using the CNN-BiLSTM competence assessment model, and output the assessment results of teachers' interdisciplinary competence by combining the matched weights and the current window dynamic threshold.

Images