Learning effect classification research method fusing BERT model and SAGEECv graph convolution

Abstract

The invention discloses a learning effect classification research method fusing BERT model and SAGEonv graph convolution. The method comprises the steps of data collection and preprocessing; collecting various kinds of process evaluation data of students in a certain period of time, and arranging the data into a format suitable for model input; constructing and training a model; the pre-trained BERT model and the SAGEEConv graph convolutional network are fused, and a classification model LABTG is constructed; performing hyper-parameter optimization; the performance of the model is optimized by adjusting hyper-parameters of the model, so that the optimal classification prediction effect is achieved; evaluating and applying the model; and using an independent test set to evaluate the classification prediction performance of the model, and according to teaching evaluation data of different stages, assisting a teacher to estimate the learning effect category of the student at the end of the stage. According to the invention, the process evaluation data of students can be analyzed more comprehensively, and the accuracy of learning effect classification prediction is improved; and real-time teaching effect feedback is provided for teachers, so that accurate teaching is realized.

Description

Technical Field

[0001] The present invention relates to the fields of educational big data analysis and machine learning, and in particular to a learning achievement classification research method integrating a BERT model and a SAGEConv graph convolution. Background Art

[0002] With the rapid development of information technology, especially the widespread application of the Internet and artificial intelligence technology, a large amount of process evaluation data has been accumulated in the field of education, such as students' homework scores, classroom test scores, laboratory report scores, and final test scores. These data contain rich information and are of great value for understanding and predicting students' learning outcomes.

[0003] However, how to effectively extract and utilize this data becomes a challenge. Traditional data analysis methods often rely on statistical models, which may not be able to fully capture the complex patterns and relationships in the data. In addition, educational data usually contains multiple types of information, such as text, numerical, and time series data. The diversity of this data increases the complexity of analysis. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a learning effectiveness classification research method that integrates the BERT model and SAGEConv graph convolution.

[0005] Technical solution: The present invention comprises the following steps:

[0006] (1) Data collection and preprocessing: Collect various types of process evaluation data of students over a certain period of time, organize these data into a format suitable for model input, and embed the data;

[0007] (2) Model construction and training: We integrate the pre-trained BERT model and the SAGEConv graph convolutional network to build a classification model, LABTG.

[0008] (3) Hyperparameter optimization: By adjusting the model’s hyperparameters, such as the weight ratio of the BERT module and the GCN module, the model’s performance is optimized to achieve the best classification prediction effect.

[0009] (4) Model evaluation and application: Use an independent test set to evaluate the model’s classification prediction performance, and assist teachers in estimating students’ learning outcomes at the end of the semester based on teaching evaluation data from different stages.

[0010] Furthermore, the step (1) includes: collecting various process evaluation data of students, organizing these data into a format suitable for model input, and embedding the data.

[0011] Furthermore, the process evaluation data includes homework scores, classroom test scores, laboratory report scores and final test scores.

[0012] Furthermore, the step (1) comprises:

[0013] (1.1) Export the student performance data of the relevant years from the academic affairs system, conduct a preliminary review of it, and remove some obviously erroneous data records;

[0014] (1.2) The data is formatted and converted into a unified format to facilitate model processing. At the same time, the data is embedded and the student score samples are converted into an embedded vector form that the model can process.

[0015] Furthermore, the LABTG model in step (2) can process serialized student performance data and capture potential correlations between performances.

[0016] Furthermore, the step (2) comprises:

[0017] (2.1) In terms of model architecture design, the structure of input layer, embedding layer, BERT module, GCN module and classification layer is adopted. The BERT module is responsible for encoding samples, while the GCN module is responsible for capturing the graph structure relationship between scores. The BERT model is pre-trained using a large amount of unlabeled corpus and fine-tuned on labeled teaching data.

[0018] (2.2) During the model training process, the cross entropy loss function is used to calculate the difference between the predicted results and the actual results, and the back propagation algorithm is used to optimize the model parameters. Through multiple iterative training, the model gradually learns how to predict the final learning outcomes of students based on their process evaluation data.

[0019] Furthermore, the hyperparameters of the model in step (3) include learning rate, batch size, and weight ratio of the BERT module and the GCN module.

[0020] Furthermore, in the hyperparameter optimization process, step (3) adopts a grid search or random search method to systematically traverse different hyperparameter combinations, and evaluates the performance of each combination on the validation set. Finally, the hyperparameter combination with the best performance on the validation set is selected as the final configuration of the model.

[0021] Furthermore, the step (4) comprises:

[0022] (4.1) In the model evaluation phase, the model's indicators on the test set are calculated, and error analysis is performed to understand on which types of samples the model performs poorly, and the model is further optimized accordingly;

[0023] (4.2) In the model application stage, the model is deployed in the actual teaching environment. Based on the students’ performance in the learning process, their learning outcomes are predicted in real time. The prediction results provide timely feedback to teachers, helping them adjust their teaching strategies and provide students with personalized learning guidance.

[0024] Furthermore, the indicators of the model in step (4.1) include precision, recall and F1 score on the test set.

[0025] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: by integrating the BERT model and the SAGEConv graph convolutional network, the present invention can more comprehensively analyze students' process evaluation data and improve the accuracy of classification prediction of learning outcomes; it can monitor academic performance at different stages of teaching and provide teachers with real-time feedback on teaching effects, thereby achieving precise teaching; it can provide students with personalized learning guidance and assistance, which helps to improve students' learning outcomes and education quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is the overall construction framework diagram of the present invention;

[0027] Figure 2 F1 value chart for the prediction of final grade categories at different stages. DETAILED DESCRIPTION

[0028] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0029] like Figure 1 As shown: The present invention provides a learning achievement classification prediction method based on the fusion of BERT model and SAGEConv graph convolution. This method can accurately predict students' learning achievements by deeply analyzing students' process evaluation data. The following are the specific implementation steps of this method:

[0030] (1) Data collection and preprocessing. This step is the basis of the entire prediction process. We collected student process teaching evaluation data from the "Operating System Principles" course of computer science majors in colleges and universities from 2020 to 2024. These data include homework grades, classroom test grades, laboratory report grades, and final test grades. The collected data needs to be preprocessed to meet the input requirements of the model. Preprocessing work includes data cleaning, formatting, and embedding representation to ensure the quality and consistency of the data.

[0031] (1.1) We export the student performance data of the relevant years from the academic affairs system and conduct a preliminary review to ensure the completeness and accuracy of the data. During this process, we may remove some obviously erroneous data records, such as abnormally high scores or missing records.

[0032] (1.2) We format the data and convert the grade data into a unified format, for example, convert all grades into numerical data and normalize them for easy model processing. At the same time, we also embed the data and convert the student grade samples into an embedded vector form that the model can process.

[0033] (2) Model construction and training: This step is the key to predicting learning outcomes. We combined the pre-trained BERT model and the SAGEConv graph convolutional network to build a classification model, LABTG. This model can process serialized student score data and capture the potential correlation between scores.

[0034] (2.1) In the design of the model architecture, we adopted the structure of input layer, embedding layer, BERT module, GCN module and classification layer. The BERT module is responsible for encoding the samples, while the GCN module is responsible for capturing the graph structure relationship between the scores. We used a large amount of unlabeled corpus to pre-train the BERT model and fine-tuned it on labeled teaching data.

[0035] (2.2) During the model training process, we used the cross entropy loss function to calculate the difference between the predicted results and the actual results, and used the back propagation algorithm to optimize the model parameters. Through multiple iterations of training, the model gradually learned how to predict the final learning outcomes of students based on their process evaluation data.

[0036] (3) Hyperparameter optimization: This step is to further improve the prediction performance of the model. We find the best model configuration by adjusting the model’s hyperparameters, such as learning rate, batch size, and the weight ratio of the BERT module and the GCN module.

[0037] In the process of hyperparameter optimization, we used grid search or random search to systematically traverse different hyperparameter combinations and evaluate the performance of each combination on the validation set. Finally, we selected the hyperparameter combination that performed best on the validation set as the final configuration of the model.

[0038] (4) Model evaluation and application. This step is an important part of testing the model’s predictive performance. We use an independent test set to evaluate the model’s classification prediction performance and assist teachers in estimating students’ learning outcomes at the end of the semester based on teaching evaluation data from different stages.

[0039] (4.1) In the model evaluation phase, we calculated the model's precision, recall, and F1 score on the test set, such as Figure 2The following figure shows the F1 value graph of the final grade category prediction at different stages to comprehensively evaluate the performance of the model. In addition, we also conducted an error analysis to understand on which types of samples the model performs poorly and further optimize the model accordingly.

[0040] (4.2) In the model application stage, we deploy the model in the actual teaching environment and predict the learning outcomes of students in real time based on their performance during the learning process. This prediction result can provide teachers with timely feedback, help them adjust their teaching strategies, and provide personalized learning guidance for students.

[0041] Example

[0042] A specific embodiment of the present invention is to apply the method of the present invention in a computer science major in a medium-sized university. In this case, we collected all the process evaluation data of 300 students in the "Operating System Principles" course and used the method of the present invention to analyze and predict it. The experimental results show that the method of the present invention can predict students' learning outcomes with a high accuracy rate, providing valuable reference information for teachers.

[0043] In this embodiment, we first preprocessed the collected data, including data cleaning, formatting and embedding representation. Then, we built the LABTG model using BERT and SAGEConv and trained it on the training set. After the model training was completed, we performed hyperparameter optimization on the validation set and finally evaluated the model on the test set. The evaluation results show that the method of the present invention has high accuracy and reliability in predicting students' learning outcomes.

[0044] Another embodiment of the present invention is to apply the method of the present invention in multiple different courses and majors. We found that despite the differences between different courses and majors, the method of the present invention can still provide accurate learning effect predictions. This shows that the method of the present invention has good generalization ability and can be widely used in different teaching scenarios.

Claims

1. A learning achievement classification research method integrating BERT model and SAGEConv graph convolution, characterized in that: The steps include: (1) Data collection and preprocessing: Collect various types of process evaluation data of students over a certain period of time, organize these data into a format suitable for model input, and embed the data; (2) Model construction and training: We integrate the pre-trained BERT model and the SAGEConv graph convolutional network to build a classification model, LABTG. (3) Hyperparameter optimization: By adjusting the model’s hyperparameters, such as the weight ratio of the BERT module and the GCN module, the model’s performance is optimized to achieve the best classification prediction effect. (4) Model evaluation and application: Use an independent test set to evaluate the model’s classification prediction performance, and assist teachers in estimating students’ learning outcomes at the end of the semester based on teaching evaluation data from different stages.

2. The learning achievement classification research method integrating BERT model and SAGEConv graph convolution according to claim 1 is characterized in that: The step (1) includes: collecting various process evaluation data of students, organizing the data into a format suitable for model input, and embedding the data.

3. The learning achievement classification research method integrating BERT model and SAGEConv graph convolution according to claim 2 is characterized in that: The process evaluation data includes homework scores, class test scores, lab report scores and final test scores.

4. The learning achievement classification research method integrating BERT model and SAGEConv graph convolution according to claim 1 is characterized in that: The step (1) comprises: (1.1) Export the student performance data of the relevant years from the academic affairs system, conduct a preliminary review of it, and remove some obviously erroneous data records; (1.2) The data is formatted and converted into a unified format to facilitate model processing. At the same time, the data is embedded and the student score samples are converted into an embedded vector form that the model can process.

5. The learning achievement classification research method integrating BERT model and SAGEConv graph convolution according to claim 1, characterized in that: The LABTG model in step (2) can process serialized student performance data and capture the potential correlation between performances.

6. The learning achievement classification research method integrating BERT model and SAGEConv graph convolution according to claim 1, characterized in that: The step (2) comprises: (2.1) In terms of model architecture design, the structure of input layer, embedding layer, BERT module, GCN module and classification layer is adopted. The BERT module is responsible for encoding samples, while the GCN module is responsible for capturing the graph structure relationship between scores. The BERT model is pre-trained using a large amount of unlabeled corpus and fine-tuned on labeled teaching data. (2.2) During the model training process, the cross entropy loss function is used to calculate the difference between the predicted results and the actual results, and the back propagation algorithm is used to optimize the model parameters. Through multiple iterative training, the model gradually learns how to predict the final learning outcomes of students based on their process evaluation data.

7. The learning achievement classification research method integrating BERT model and SAGEConv graph convolution according to claim 1, characterized in that: The hyperparameters of the model in step (3) include learning rate, batch size, and weight ratio of the BERT module and the GCN module.

8. The learning achievement classification research method integrating BERT model and SAGEConv graph convolution according to claim 1, characterized in that: In the hyperparameter optimization process, step (3) uses a grid search or random search method to systematically traverse different hyperparameter combinations, and evaluates the performance of each combination on the validation set. Finally, the hyperparameter combination with the best performance on the validation set is selected as the final configuration of the model.

9. The method for classifying learning outcomes by integrating the BERT model and SAGEConv graph convolution according to claim 1, characterized in that: The step (4) comprises: (4.1) In the model evaluation phase, the model's indicators on the test set are calculated, and error analysis is performed to understand on which types of samples the model performs poorly, and the model is further optimized accordingly; (4.2) In the model application stage, the model is deployed in the actual teaching environment. Based on the students’ performance in the learning process, their learning outcomes are predicted in real time. The prediction results provide timely feedback to teachers, helping them adjust their teaching strategies and provide students with personalized learning guidance.

10. The method for learning effectiveness classification based on the fusion of BERT model and SAGEConv graph convolution according to claim 9, characterized in that: The indicators of the model in step (4.1) include precision, recall and F1 score on the test set.

Images