Student meta-cognitive level prediction method, system and equipment

Abstract

The invention belongs to the technical field of computer application, and particularly discloses a student meta-cognition level prediction method, system and equipment, and the method comprises the steps: obtaining a static variable reflecting the student meta-cognition level; the static variables comprise a plurality of variables belonging to student basic information, student family information, student school states and student character corresponding variables; inputting the static variable into a pre-trained prediction model to obtain a prediction result of the meta-cognitive level of the student; the prediction model comprises a feature extraction module and a meta-cognitive prediction module; the feature extraction module is used for extracting features related to the meta-cognition level in combination with the incidence relation among all the static variables; the meta-cognitive prediction module is used for obtaining a meta-cognitive level prediction result based on the obtained features; the feature extraction module comprises a bidirectional LSTM (Long Short Term Memory) added with Peephole connection, and the Peephole connection directly observes each static variable through a cell state. According to the method and the device, the accuracy of meta-cognitive level prediction is improved.

Description

Technical Field

[0001] The present application belongs to the field of computer application technology, and more specifically, relates to a method, system and device for predicting students' metacognitive level. Background Art

[0002] The number of metacognition-related papers published in top international journals is on the rise, and metacognition research covers a wide range of topics. Current research focuses on exploring the elements of the underlying theory, investigating its practical online applications, and analyzing its effectiveness. The findings are expected to inform decisions about potential areas for future research.

[0003] International research indicates that predicting students' metacognitive abilities has become a hot research area. Current research on predicting students' metacognitive abilities focuses on measuring them using various assessment tools and methods. These tools and methods, including questionnaires, cognitive tasks, and self-reports, can assess students' understanding of their learning processes and their ability to use and adjust learning strategies. Furthermore, some studies have explored the correlation between metacognitive abilities and academic achievement, learning strategies, and the impact of metacognitive training on student learning outcomes. These findings provide valuable insights for educational practice and individual student development.

[0004] Some international studies primarily use interviews and self-assessments to measure metacognitive skills. However, the data collected in these methods is highly subjective, potentially inaccurate, and can only measure learners' metacognitive skills to a limited extent. Gloria A. Stillman and Peter L. Galbraith conducted an in-depth study of high school girls' problem-solving activities. They used answer graphs to analyze and categorize individual students' written responses. They then used videotaped conferences and structured free-response interviews to examine students' metacognitive knowledge, strategies, decision-making, beliefs, and influences.

[0005] To make research more objective, researchers have begun developing questionnaires and scales related to metacognition. Several well-known metacognition questionnaires and scales exist internationally, including the Metacognitive Awareness Inventory (MAI), the Metacognitive Awareness of Reading Strategies Inventory (MARSI), and the Metacognitive Questionnaire for Children (MQC). These scales and questionnaires are widely used to assess students' metacognitive proficiency, helping researchers and educators better understand students' understanding of their own cognitive processes and their use of metacognitive strategies in learning. These tools help reveal students' metacognitive proficiency, providing important insights for personalized instruction and learning strategy guidance. Schraw and Sperling-Denisson developed a 52-item Metacognitive Awareness Inventory (MAI) to measure metacognitive awareness in adults. Rayne A. Sperling, Bruce C. Howard, and others measured children's metacognition, discussing factors influencing students' metacognitive proficiency and exploring gender and grade differences. Their scale has demonstrated high validity.

[0006] Although existing technology can assess students' metacognitive proficiency through metacognitive questionnaires, existing research on predicting metacognitive proficiency faces several challenges: 1. Lack of objectivity in data: Existing research methods primarily focus on subjective attributes, such as students' self-evaluations of their learning process. This approach ignores static student attributes, including their parents' education level, schooling methods, attendance rates, and health status. This emphasis on subjective attributes in model construction results in an incomplete understanding of the student's overall situation, which in turn affects the accuracy and reliability of the predictive model. 2. Inefficient model learning: Most predictive methods rely on traditional machine learning algorithms, which struggle to capture the nonlinear relationships and interactions inherent in complex learning processes. Furthermore, these algorithms have high requirements for data quality and quantity, making them susceptible to noise and sample bias. As a result, predictive models often suffer from overfitting or underfitting, reducing prediction accuracy. 3. Complexity of the prediction process: Currently, there is a lack of online platforms that can quickly and accurately predict mathematical metacognitive proficiency. Existing prediction methods typically rely on questionnaires, which are time-consuming and complex to collect. The lack of easy-to-use tools and platforms limits the ability of researchers and educators to effectively predict and apply mathematical metacognitive proficiency. Summary of the Invention

[0007] In response to the shortcomings of the existing technology, the purpose of this application is to provide a method, system and equipment for predicting students' metacognitive level, aiming to solve the problems of limited accuracy of existing student metacognitive level prediction schemes and the inability to make quick and accurate predictions.

[0008] To achieve the above objectives, in a first aspect, the present application provides a method for predicting a student's metacognitive level, comprising: Obtaining static variables reflecting the student's metacognitive level; the static variables include: multiple variables belonging to student basic information, student family information, student school status, and student character corresponding variables; The static variables are input into a pre-trained prediction model to obtain a prediction result of the student's metacognitive level; the prediction model includes: a feature extraction module and a metacognitive prediction module; the feature extraction module extracts features related to the metacognitive level from the correlation relationships among all static variables; the metacognitive prediction module obtains a metacognitive level prediction result based on the acquired features; the feature extraction module includes: a bidirectional long short-term memory network (Long Short-Term Memory, LSTM) with a peephole connection added, the Peephole connection directly observes each static variable through the cell state, so that the LSTM can maintain the long-term influence of important static variables.

[0009] In a possible implementation, the metacognitive prediction module includes: a multilayer perceptron (MLP); The MLP includes: three fully connected layers; the first fully connected layer and the second fully connected layer are each connected to a dropout operation and a ReLU activation function; the dropout operation randomly sets the output of some neurons to 0 during training to reduce the complexity of the metacognitive prediction module, and the ReLU activation function is used to enhance the nonlinear expression ability of the metacognitive prediction module; a Sigmoid activation function is connected after the third fully connected layer to produce the final prediction result.

[0010] In a possible implementation, the prediction model further includes: a word vector embedding module, which is used to convert all static attributes into digital feature vectors and then output them to the feature extraction module.

[0011] In one possible implementation, the label values ​​used in the prediction model pre-training process are obtained based on a metacognitive questionnaire filled out by students.

[0012] In a possible implementation, the variables corresponding to the student's basic information include: at least one of: gender, whether the student is an only child, and the ranking of non-only children; The variables corresponding to the student's family information include: at least one of the father's education level, the mother's education level, the father's occupation, and the mother's occupation; The variables corresponding to the student's school attendance status include: at least one of health status, school attendance mode, and attendance; The variables corresponding to the student's character include: personality, information collection method, thinking method, execution method, belief, emotional stability, interest, self-efficacy, classroom concentration, divergent thinking ability and learning communication ability.

[0013] In one possible implementation, the static variables are ranked in descending order of importance, including: emotional stability, classroom concentration, divergent thinking ability, self-efficacy, learning and communication skills, whether or not the student is an only child, school mode, attendance, execution method, and information collection method.

[0014] In a second aspect, the present application provides a student metacognitive level prediction system, comprising: A variable acquisition unit is used to acquire static variables reflecting the student's metacognitive level; the static variables include: multiple variables corresponding to the student's basic information, student's family information, student's school status, and student's character; A prediction model is provided for obtaining a prediction result of a student's metacognitive level based on the static variables; the prediction model comprises: a feature extraction module and a metacognitive prediction module; the feature extraction module extracts features related to the metacognitive level from the static variables in combination with the correlation relationships among all the static variables; the metacognitive prediction module obtains a metacognitive level prediction result based on the acquired features; the feature extraction module comprises: a bidirectional LSTM with a peephole connection added, the peephole connection directly observing each static variable through the cell state, so that the LSTM can maintain the long-term influence of important static variables.

[0015] In one possible implementation, the metacognitive prediction module includes: an MLP; the MLP includes: three fully connected layers; a dropout operation and a ReLU activation function are each connected after the first fully connected layer and the second fully connected layer; the dropout operation randomly sets the output of some neurons to 0 during the training process to reduce the complexity of the metacognitive prediction module, and the ReLU activation function is used to enhance the nonlinear expression ability of the metacognitive prediction module; a Sigmoid activation function is connected after the third fully connected layer to generate the final prediction result.

[0016] In one possible implementation, the variable acquisition unit acquires static variables, and the variables corresponding to the student's basic information include: gender, whether the child is an only child, and at least one of the ranking of non-only children; the variables corresponding to the student's family information include: at least one of the father's education level, the mother's education level, the father's occupation, and the mother's occupation; the variables corresponding to the student's school status include: at least one of health status, school mode, and attendance; and the variables corresponding to the student's character include: personality, way of collecting information, way of thinking, way of execution, belief, emotional stability, interest, self-efficacy, classroom concentration, divergent thinking ability, and learning communication ability.

[0017] In a third aspect, the present application provides an electronic device comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0019] In a fifth aspect, the present application provides a computer program product, which, when executed on a processor, enables the processor to execute the method described in the first aspect or any possible implementation of the first aspect.

[0020] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the existing technologies: This application provides a method, system, and device for predicting a student's metacognitive level. This application provides a method, system, and device for predicting a student's metacognitive level. By using static attributes to predict metacognitive levels, the student's true metacognitive level can be directly derived through the model. Compared to questions in a metacognitive monitoring questionnaire, these static attributes are less susceptible to subjective influences from students and can more accurately reflect their metacognitive level. Verification has shown that the student metacognitive monitoring prediction model proposed in this application outperforms traditional machine learning models in terms of precision, recall, and F1 value. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of the process of predicting students' metacognitive level provided by the embodiment of the present application; Figure 2 is a schematic diagram of the prediction model provided in the embodiment of the present application; Figure 3This is a comparison histogram of the SMML-Net model provided in the embodiment of the present application and the machine learning evaluation index; Figure 4 This is a comparison chart of the ablation experiment provided in the embodiment of the present application on the evaluation indicators of ACC, F1 score, precision and recall rate; Figure 5 This is a comparison chart of the RMSE evaluation index of the ablation experiment provided in the embodiment of the present application; Figure 6 This is a flow chart of feature importance evaluation provided by an embodiment of the present application; Figure 7 This is an architecture diagram of a student metacognitive level prediction system provided by an embodiment of the present application; Figure 8 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0023] The term "and / or" in this application describes an association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, or B exists alone. The symbol " / " in this application indicates that the associated objects are in an "or" relationship, for example, A / B means A or B.

[0024] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0025] In the description of the embodiments of the present application, unless otherwise specified, “at least one” means one or more than one, for example, at least one variable means one or more variables, etc.

[0026] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.

[0027] Figure 1 This is a flow chart of the method for predicting students' metacognitive level provided by the embodiment of the present application; Figure 1 As shown, the following steps are included: Step S101, obtaining static variables reflecting the student's metacognitive level; the static variables include: multiple variables corresponding to the student's basic information, student's family information, student's school status, and student's character; For example, the above static variables (also called static features) can be found in Table 1:

[0028] The above static variables can be obtained through questionnaires or by third parties (such as teachers).

[0029] Step S102: Input the static variables into a pre-trained prediction model to obtain a prediction result of the student's metacognitive level; the prediction model includes: a feature extraction module and a metacognitive prediction module; the feature extraction module extracts features related to the metacognitive level from the correlation relationships among all static variables; the metacognitive prediction module obtains a metacognitive level prediction result based on the acquired features; the feature extraction module includes: a bidirectional LSTM with a peephole connection, the Peephole connection directly observes each static variable through the cell state, so that the LSTM can maintain the long-term influence of important static variables.

[0030] It's important to note that among the static variables mentioned above, personal factors (gender, personality, and self-efficacy) influence cognitive strategy selection; environmental factors (family background, schooling style) indirectly shape metacognition through resource acquisition and social support; and behavioral factors (classroom focus and attendance) reflect the explicit manifestations of metacognition. Self-efficacy, as a core variable, directly relates to metacognition's ability to judge task difficulty and adjust strategies. Static variables such as "thinking style (thinking / feeling)" and "information gathering style (sensing / intuitive)" reflect individual information processing preferences and directly influence the monitoring and regulation mechanisms in metacognition. For example, intuitive students may be more adept at optimizing learning paths through abstract connections. Family variables (parents' education level and occupation) belong to the microsystem and influence metacognitive development through cultural capital; boarding / day-schooling patterns belong to the mesosystem, reflecting the interactive effects of family and school, which may influence opportunities for independent learning. Variables such as extroversion / introversion and emotional stability are related to the "planning-monitoring-evaluation" cycle of metacognition. For example, extroverted students may be more proactive in using metacognitive strategies in collaborative learning.

[0031] In summary, the method of using the above 21 static variables to directly predict students' metacognitive level can integrate theories such as social cognition and ecological systems, break through the single perspective limitations of traditional questionnaires, and provide a more objective and systematic metacognitive analysis framework.

[0032] The purpose of this application model design is to predict students' mathematical metacognitive level by combining their static features with various mathematical ability characteristics. The deep learning-based SMML-Net model (i.e., the aforementioned prediction model) greatly simplifies its use for researchers. Simply by providing information about the student's static attributes, the model can directly determine the student's true metacognitive level. This allows students to learn according to the targeted learning model proposed in this study, thereby helping parents and teachers improve student performance.

[0033] Specifically, the entire prediction model framework consists of three parts: Figure 2 As shown in the figure, the first part is the word embedding module, the second part is the feature extraction module, and the third part is the metacognitive prediction module. The word embedding module aims to encode the sample's attribute features into digital features for subsequent processing. The feature extraction module aims to identify correlations between the sample's numerous attribute features and fuse them. The metacognitive prediction module learns the extracted features and ultimately provides a metacognitive prediction result for the sample.

[0034] More specifically, 1. For the word vector embedding module: After completing the screening of the static attributes of the sample, the first task is to convert these attributes into digital feature vectors that the model can effectively process. In this conversion process, this study adopted a feature conversion strategy that regards each independent attribute as a "vocabulary" and all the attributes of a sample as constituting a "sentence". For example, a student's multi-dimensional attributes can be compared to the vocabulary in a sentence, such as gender, whether they are an only child, self-efficacy, etc., which together form a comprehensive description of the student. This method makes the extraction of features from complex data more detailed and systematic, which helps to more accurately capture the subtle differences that affect students' metacognitive monitoring levels in subsequent analysis.

[0035] 2. For the feature extraction module: In this study, the feature extraction process plays a crucial role in parsing the embedding matrix that constitutes the student features. The research aims to guide the model to move beyond relying on single "vocabulary" features and instead accurately determine the metacognitive monitoring level of students by comprehensively evaluating the overall information of students, that is, adopting a feature extraction method based on the complete "sentence" context. Bidirectional LSTM + Peephole connections are used to enhance the ability to handle long-term dependencies in memory. In the bidirectional LSTM, the forward propagation can establish the influence path of "feature i → feature j" (i < j), and the backward propagation can establish the influence path of "feature j → feature i" (i < j). For example, assuming the feature arrangement order: 17: self-efficacy, 18: classroom concentration. At the psychological level, in the forward influence, high self-efficacy can enhance classroom concentration, and in the backward influence, continuous classroom concentration performance can enhance self-efficacy. Through this bidirectional structure design, the model can not only capture traditional forward causal relationships (such as "high self-efficacy leads to better classroom concentration"), but also discover reverse influence paths (such as "continuous improvement in classroom concentration in turn enhances self-efficacy"). This modeling ability of bidirectional interaction makes the model have stronger interpretability and prediction accuracy for the complex interaction of psychological features in educational scenarios.

[0036] The Peephole connection is an improved structure of LSTM that allows the gating mechanism to directly access the cell state, thus more precisely controlling the flow of information. A student's "self-efficacy" (feature 17) may continuously affect multiple subsequent features (such as "classroom concentration" 18, "learning communication ability" 19). Through the direct observation of the cell state, the Peephole connection enables the model to more accurately maintain the long-term influence of important features. For example, when the cell state stores the intensity information of "self-efficacy": the input gate can decide whether to accept new relevant features (such as changes in "classroom concentration") based on and the forget gate can judge when to weaken the outdated psychological state information based on

[0037] Forward propagation path: Feature 17 (self-efficacy) → [Peephole adjustment] → Feature 18 (classroom concentration).

[0038] Backward propagation path: Feature 18 (classroom concentration) → [Peephole adjustment] → Feature 17 (self-efficacy).

[0039] Key information transmitted through the cell state: C_17 contains the strength of self-efficacy → affects the gated calculation of feature 18; or C_18 contains the level of classroom concentration → reversely affects the gated update of feature 17.

[0040] Furthermore, educational data typically has limited sample sizes and is prone to overfitting. Peephole connections constrain cell states, making gating decisions more logical (e.g., "when self-efficacy is sufficiently high, this feature is no longer easily forgotten") and reducing the impact of random noise.

[0041] 3. For the metacognitive prediction module: Through the feature extraction module, after evaluating the student digital portrait of each student, a feature expression vector for each student can be obtained. It is a digital feature composed of a one-dimensional vector. The digital feature is then input into a multi-layer perceptron (MLP) for further processing. The multi-layer perceptron is a feedforward neural network consisting of an input layer, a hidden layer, and an output layer. In the embodiment of the present application, the multi-layer perceptron consists of three fully connected layers, which are connected by a weight matrix and a bias vector. The output of each fully connected layer is the weighted sum of the input features, which has undergone a nonlinear transformation of the activation function.

[0042] The first two fully connected layers are followed by a dropout operation and a ReLU activation function. The ReLU activation function is specifically: . Where x represents the weighted sum of all inputs of the previous layer plus the bias.

[0043] Dropout is a regularization technique that effectively reduces model complexity and prevents overfitting by randomly setting the output of some neurons to 0 during training. The ReLU activation function is a commonly used nonlinear activation function that outputs all negative values ​​as 0 and retains all positive values, increasing the model's nonlinear expressiveness.

[0044] The last fully connected layer uses the Sigmoid activation function, the formula is , used to generate the final prediction result. The Sigmoid function maps the output value to the range (0, 1), which can be interpreted as a probability value. In this work, the output of the activation function is the probability of the sample being at high, medium, or low metacognitive levels. Ultimately, the highest probability value is selected as the predicted metacognitive level of the sample.

[0045] In general, the multilayer perceptron increases the depth and nonlinear expression ability of the model by stacking multiple fully connected layers and activation functions, so that it can more effectively learn complex patterns from features and improve prediction accuracy.

[0046] In summary, this application aims to achieve rapid prediction of students' metacognitive monitoring level by establishing a deep learning model based on student static data prediction.

[0047] In one embodiment, to address the shortcomings of current research, this application collected 1,985 student data sets to expand the existing student-related datasets, and predicted students' metacognitive monitoring levels based on a hybrid neural network model, improving the accuracy of the prediction. The specific research content is as follows: 1. Using the Wenjuanxing platform as a testing tool, we distributed questionnaires on students' metacognitive monitoring levels. We selected students from middle schools in Tai'an, Shandong Province as the test group. We planned the overall student population, determined the schools to be measured, and the number of students to be measured in each school. 2. Organize the returned questionnaires and screen out invalid ones. Use SPSS software to conduct reliability, validity and stability analysis on the measured data; 3. Construct a student data set and classify the standards for students' metacognitive monitoring levels, clarify the characteristics of students at each level, normalize each static feature value, delete abnormal sample data, and encode the attribute feature values; 4. Construct the SMML-Net model. First, use the traditional machine learning algorithms of random forest, decision tree algorithm, gradient boosting decision tree algorithm, K nearest neighbor, naive Bayes and support vector machine to establish single models to predict students' metacognitive monitoring level. To further improve the model prediction effect, a neural network model is constructed. The student metacognitive monitoring prediction model proposed in this application is superior to the traditional machine learning model in terms of precision, recall rate and F1 value. Then select the ten most important static variables for predicting students' metacognitive monitoring level; 5. Based on in-depth data analysis and chi-square tests, identify key factors that significantly influence students' metacognitive monitoring levels. Based on the differences in the characteristics of different students' metacognitive monitoring levels and actual teaching situations, provide constructive suggestions. 6. Build a platform for predicting students' metacognitive monitoring levels. This system was developed using Java and Python, along with front-end and back-end development frameworks and deep learning frameworks. The system accurately predicts students' metacognitive monitoring levels by selecting ten required static data points. Batch predictions can be performed by uploading files, and learning plans tailored to different metacognitive monitoring levels can be viewed on the download interface.

[0048] The specific implementation steps are as follows: 1. Filter data: The dataset for this application consists of 1,985 student questionnaires and data provided by an online platform used by all students. After conducting validity and reliability analyses to verify the credibility of the questionnaire, we retained 18 static student attributes and three teacher-assessed student competency attributes (classroom focus, divergent thinking, and learning communication). To ensure the credibility of the 1,985 sample data, we used the time students spent completing the questionnaire to assess its quality.

[0049] This application believes that if a student takes too little time to complete a questionnaire, it may indicate that they haven't thought carefully about the content. If it takes too long, it may be because they were distracted or had something else happen. These factors can significantly affect the quality of the questionnaire and cause the results to deviate significantly from the student's actual situation. Therefore, we processed the questionnaire duration using a boxplot. After filtering, 1816 samples were retained, with an efficiency of approximately 91.5%.

[0050] 2. Metacognitive Monitoring Level Label: In this application, 30 questions in a conventional metacognitive monitoring questionnaire (Likert self-report scale) are used, as shown in Table 2. Table 2 is a modified general metacognitive questionnaire with high credibility, which is used to obtain the label value corresponding to each sample.

[0051] Table 2 Metacognition Questionnaire

[0052] The sample's metacognitive level was assessed. Areti Panaoura, Philippou, and others believe that higher scores on questions about metacognitive level correspond to higher metacognitive levels. Therefore, metacognitive levels are divided into three levels: high, medium, and low. Unlike other studies, this study used K-means unsupervised clustering of the total score T of the 30 questionnaire questions, clustering the sample into three clusters.

[0053] 3. Input the above static variables into the constructed prediction model to predict the level of metacognitive monitoring.

[0054] 4. Evaluate the importance of features based on the model’s results: In daily life, parents or teachers may not be able to quickly provide 21 static attributes and mathematical ability characteristics of students, which may take a considerable amount of time. To simplify the process of users predicting the level of mathematical metacognitive monitoring, this work studied the importance and relevance of these 21 different characteristics, screened the 10 most important characteristics, and used these 10 characteristics to predict the level of metacognitive monitoring of high school students.

[0055] This application first scrambles the initial features, that is, the initial feature values ​​are randomly exchanged and have the same distribution as the original attributes. Represents the i-th initial feature, and Indicates the Then, each feature is shuffled in turn and input into the SMML-Net model together with other initial features, and the new SMML-Net model is trained to calculate the performance of the model in the test set. Evaluation indicators, using Indicates the evaluation index trained in the SMML-Net model using the original features, Represents the evaluation index after disrupting the i-th feature. The importance of each feature is expressed by the following formula:

[0056] 5. Propose learning improvement plans for students with different levels of metacognitive monitoring: In this application, the goal is to improve students' metacognitive monitoring, so the focus is on student characteristics and abilities that can be shaped and improved by subsequent learning plans, parental involvement, and school intervention. Through a detailed chi-square test, the goal is to identify key factors that have a significant impact on students' metacognitive monitoring. These factors may include but are not limited to students' learning habits, emotional management skills, thinking styles, and classroom participation. The results of the statistical analysis revealed the correlation between these variables and the level of metacognitive monitoring, providing data support for educational intervention.

[0057] In particular, the application of the chi-square test not only helps quantify the influence of each factor, but also points out those areas with potential influence. For example, when a feature shows a high statistical significance in the chi-square test, it can be reasonably inferred that this feature is closely related to the development of metacognitive monitoring level. Such findings provide valuable insights for educators, allowing them to be more targeted when developing personalized teaching strategies and interventions. In addition, by comparing the different features in the chi-square test, values, can determine which variables are more statistically important, low A value of means that the association of a trait with the level of metacognitive monitoring is unlikely to occur by chance, and those with higher The characteristics of the values ​​may require further investigation, as their effects may differ across educational backgrounds or demographics.

[0058] 6. Design a rapid prediction platform for metacognitive monitoring levels: To better apply the SMML-Net model, this work built a platform for rapid prediction of metacognitive monitoring levels. The platform's frontend is built using the Vue framework, and the backend is implemented using the Spring Boot + Flask framework. The prediction platform is constructed by invoking the Flask service interface through the Spring Boot framework's HttpServlet. The following section will discuss requirements analysis, platform design, and technical implementation.

[0059] The task of predicting high school students' metacognitive proficiency is a three-class supervised learning task, so many evaluation metrics used for binary prediction require refinement to be applicable. In this study, the SMML-Net model was evaluated using macro-averaged precision, recall, and F1 scores, as well as multi-class accuracy and root mean square error (RMSE).

[0060] Macro-averaging refers to the process of simplifying a multi-classification task into multiple binary-classification tasks when calculating its evaluation indicators, calculating the evaluation indicators of the binary-classification tasks (such as precision and recall), and then averaging the indicators of all categories. Macro-averaging gives each category the same weight and is suitable for situations where the labels of each category are equally important. In this study, the three labels correspond to high, medium, and low metacognitive levels, and their importance is equal. The number of the three labels is 366, 821, and 628, respectively, and the number of the three labels is relatively balanced. The formula for macro-averaging is as follows:

[0061] in, is the total number of categories, It is The evaluation indicators of each category, It is a macro average. Macro average does not favor the category with more samples due to the different number of category samples. It focuses more on the overall performance of each category.

[0062] Accuracy It is the simplest and most intuitive observation indicator in classification problems, but it has a fatal disadvantage: when the distribution of positive and negative samples is uneven, the accuracy has almost no observation ability. The reason is that when the negative samples account for 99%, even if all the results are predicted as negative, the accuracy rate can still reach 99%. It refers to the ratio of the number of correctly classified positive samples to the number of positive samples determined by the classifier. It can be understood as how many of the positive samples are predicted correctly. It also refers to the proportion of samples that are predicted to be positive in the true positive samples. F1 score Also known as the balanced F score, it is defined as the harmonic mean of precision and recall, which can more intuitively observe the average performance of a model. This is a key evaluation metric in multi-classification tasks, representing the average deviation between the predicted and true values. In prediction tasks, completely accurate predictions are often impossible, so researchers prefer predictions that are closer to the true results. Specifically, this work aims to minimize the number of students with high metacognitive monitoring levels being predicted as having low metacognitive levels, which would significantly reduce the effectiveness of targeted learning strategies.

[0063] To explore the effectiveness of the SMML-Net model, this application compared six classic machine learning methods: decision tree (DT), gradient boosted decision tree (GBDT), nearest neighbor (KNN), naive Bayes (NB), random forest (RF), and support vector machine (SVM). This study employed a 5-fold cross-validation approach, splitting the training set into five parts, with four parts used as training sets each time and the remaining part used as a test set. This yielded five training / test set combinations, resulting in five training and testing cycles. Evaluation metrics were all averaged across the five training and testing cycles, ensuring more reliable performance estimates.

[0064] like Figure 3 As shown in Table 3, the SMML-Net model using RNNs has a clear advantage in accuracy, F1 score, precision, and recall. This demonstrates that the strategy of treating sample attributes as a sentence is feasible and efficient. Secondly, SVMs, due to their superior capabilities in high-dimensional spaces and small sample datasets, also achieved relatively strong performance in this research. As shown in Table 3, the SMML-Net model outperformed the SVM model by 4.40%, 4.38%, 5.03%, and 3.51% in accuracy, F1 score, precision, and recall, respectively.

[0065] Table 3 Average evaluation indicators of SMML-Net model and machine learning methods

[0066] In the multi-classification task, the SMML-Net model also has excellent performance in the very important root mean square error. Figure 3 It can be seen that the SMML-Net model performed stably during the five training and testing processes, without any outliers, indicating that the experimental data set division was scientific and random.

[0067] Using the SMML-Net model to process students' static attributes and mathematical abilities is crucial for the neural network to be able to effectively predict students' mathematical metacognitive monitoring levels. To this end, while keeping all hyperparameters unchanged, we designed a feature extraction module using three feature extraction networks: CNN, LSTM, and self-attention. We conducted an ablation experiment. The specific results are shown in [1]. Figure 4 and Figure 5 .

[0068] like Figure 4 and Figure 5 As shown, the SMML-Net model has excellent performance in predicting tasks with shorter sentences and smaller data sets. Although RNN often has problems with gradient vanishing or gradient exploding when processing longer sentences, the hidden state of RNN can retain historical information to a certain extent and also has good memory capabilities. At the same time, compared with the LSTM model and the Self-attention model, the RNN model has fewer parameters and is simpler, and is less likely to overfit when the data set is small. The CNN model also has fewer parameters, but the CNN model cannot capture the dependencies between contexts when understanding long sentences and is not suitable for the research of this application.

[0069] In daily life, parents or teachers may not be able to quickly provide 21 static attributes and mathematical ability characteristics of students, which may take a considerable amount of time. To simplify the process of users predicting the level of mathematical metacognitive monitoring, this work studied the importance and relevance of these 21 different characteristics, screened the 10 most important characteristics, and used these 10 characteristics to predict the level of metacognitive monitoring of high school students.

[0070] like Figure 6 As shown, the embodiment of the present application first scrambles the initial features, that is, the initial feature values ​​are randomly exchanged and have the same distribution as the original attributes. Representative Initial features, using Indicates the Then, each feature is disrupted in turn and input into the SMML-Net model together with other initial features, and the RMSE evaluation index of the model trained by the new SMML-Net model in the test set is calculated. Indicates the evaluation index trained in the SMML-Net model using the original features, Indicates disruption of Evaluation indicators after features. Indicates the importance of each feature: Subsequently, the top ten rankings of the importance of the 21 characteristics of the students were calculated, as shown in Table 4.

[0071] Table 4 Feature importance ranking table

[0072] It can be seen that whether students' emotions are stable, classroom concentration, divergent thinking ability, self-efficacy, learning and communication skills, whether they are an only child, school attendance, decision-making methods, and information collection methods are the ten most important characteristics in predicting students' metacognitive monitoring level.

[0073] Experimental verification shows that the prediction accuracy can reach 91.3% using only ten important features, and 94.2% using all features. This shows that the embodiment of the present application can achieve highly accurate metacognitive prediction using only the above 10 static variables.

[0074] Figure 7 This is the architecture diagram of the student metacognitive level prediction system provided by the embodiment of the present application, such as Figure 7 As shown, including: The variable acquisition unit 710 is used to acquire static variables reflecting the student's metacognitive level; the static variables include: multiple variables corresponding to student basic information, student family information, student school status, and student character; Prediction model 720 is used to obtain a prediction result of the student's metacognitive level based on the static variables; the prediction model includes: a feature extraction module and a metacognitive prediction module; the feature extraction module extracts features related to the metacognitive level from the static variables based on the correlation relationship between all the static variables; the metacognitive prediction module obtains a metacognitive level prediction result based on the acquired features; the feature extraction module includes: a bidirectional LSTM with a peephole connection added, the Peephole connection directly observes each static variable through the cell state, so that the LSTM can maintain the long-term influence of important static variables.

[0075] The metacognitive prediction module includes: an MLP; the MLP includes: three fully connected layers; a dropout operation and a ReLU activation function are respectively connected after the first fully connected layer and the second fully connected layer; the dropout operation randomly sets the output of some neurons to 0 during training to reduce the complexity of the metacognitive prediction module, and the ReLU activation function is used to enhance the nonlinear expression ability of the metacognitive prediction module; a Sigmoid activation function is connected after the third fully connected layer to produce the final prediction result.

[0076] For example, the variable acquisition unit 710 acquires static variables, and the variables corresponding to the student's basic information include: gender, whether or not the child is an only child, and at least one of the ranking of non-only children; the variables corresponding to the student's family information include: at least one of the father's education level, the mother's education level, the father's occupation, and the mother's occupation; the variables corresponding to the student's school status include: at least one of health status, school mode, and attendance; and the variables corresponding to the student's character include: personality, way of collecting information, way of thinking, way of execution, belief, emotional stability, interest, self-efficacy, classroom concentration, divergent thinking ability, and learning communication ability.

[0077] It should be understood that the above-mentioned system is used to execute the methods in the above-mentioned embodiments. The implementation principles and technical effects of the corresponding program units, models and modules in the system are similar to those described in the above-mentioned methods. The working process of the system can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0078] Based on the method in the above embodiment, the embodiment of the present application provides an electronic device, such as Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the method in the above embodiment.

[0079] In addition, the logic instructions in the aforementioned memory 830 can be implemented in the form of a software functional unit and, when sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0080] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0081] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0082] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0083] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and storage medium can be located in an ASIC.

[0084] The above embodiments can be implemented in whole or in part using software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions. When loaded and executed on a computer, the computer program instructions fully or partially produce the processes or functions described in the embodiments of this application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted via the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disk, hard disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid-state drive (SSD)).

[0085] It will be understood that the various numerical numbers involved in the embodiments of the present application are merely distinctions for the convenience of description and are not intended to limit the scope of the embodiments of the present application.

[0086] It is easy for those skilled in the art to understand that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for predicting students' metacognitive level, characterized by: include: Obtain static variables that reflect students' metacognitive level; The static variables include: multiple variables corresponding to student basic information, student family information, student school status, and student character; The static variables are input into a pre-trained prediction model to obtain a prediction result of the student's metacognitive level; the prediction model includes: a feature extraction module and a metacognitive prediction module; the feature extraction module extracts features related to the metacognitive level from the correlation relationships among all static variables; the metacognitive prediction module obtains a metacognitive level prediction result based on the acquired features; the feature extraction module includes: a bidirectional long short-term memory network LSTM with a peephole connection added, the Peephole connection directly observes each static variable through the cell state, so that the LSTM can maintain the long-term influence of important static variables.

2. The method according to claim 1, characterized in that The metacognitive prediction module includes: a multi-layer perceptron MLP; The MLP includes: three fully connected layers; the first fully connected layer and the second fully connected layer are each connected to a dropout operation and a ReLU activation function; the dropout operation randomly sets the output of some neurons to 0 during training to reduce the complexity of the metacognitive prediction module, and the ReLU activation function is used to enhance the nonlinear expression ability of the metacognitive prediction module; a Sigmoid activation function is connected after the third fully connected layer to produce the final prediction result.

3. The method according to claim 1, characterized in that The prediction model also includes: a word vector embedding module, which is used to convert all static attributes into digital feature vectors and then output them to the feature extraction module.

4. The method according to any one of claims 1 to 3, characterized in that The label values ​​used in the pre-training process of the prediction model are obtained based on the metacognitive questionnaire filled out by students.

5. The method according to claim 1, wherein The variables corresponding to the student's basic information include: gender, whether the student is an only child, and at least one of the ranking of non-only children; The variables corresponding to the student's family information include: at least one of the father's education level, the mother's education level, the father's occupation, and the mother's occupation; The variables corresponding to the student's school attendance status include: at least one of health status, school attendance mode, and attendance; The variables corresponding to the student's character include: personality, information collection method, thinking method, execution method, belief, emotional stability, interest, self-efficacy, classroom concentration, divergent thinking ability and learning communication ability.

6. The method according to claim 1 or 5, characterized in that The static variables include: emotional stability, classroom concentration, divergent thinking ability, self-efficacy, learning and communication skills, whether or not one is an only child, school attendance pattern, execution method, and information collection method.

7. A student metacognitive level prediction system, characterized by: include: Variable acquisition unit, used to obtain static variables reflecting students' metacognitive level; The static variables include: multiple variables corresponding to student basic information, student family information, student school status, and student character; A prediction model is provided for obtaining a prediction result of a student's metacognitive level based on the static variables; the prediction model comprises: a feature extraction module and a metacognitive prediction module; the feature extraction module extracts features related to the metacognitive level from the static variables in combination with the correlation relationships among all the static variables; the metacognitive prediction module obtains a metacognitive level prediction result based on the acquired features; the feature extraction module comprises: a bidirectional long short-term memory network (LSTM) with a peephole connection added, the Peephole connection directly observing each static variable through the cell state, so that the LSTM can maintain the long-term influence of important static variables.

8. The system according to claim 7, characterized in that The metacognitive prediction module includes: a multi-layer perceptron (MLP); the MLP includes: three fully connected layers; a dropout operation and a ReLU activation function are respectively connected after the first fully connected layer and the second fully connected layer; the dropout operation randomly sets the output of some neurons to 0 during the training process to reduce the complexity of the metacognitive prediction module, and the ReLU activation function is used to enhance the nonlinear expression ability of the metacognitive prediction module; a Sigmoid activation function is connected after the third fully connected layer to generate the final prediction result.

9. The system according to claim 7, wherein: The variable acquisition unit acquires static variables, wherein the variables corresponding to the student's basic information include: gender, whether the student is an only child, and at least one of the ranking of non-only children; the variables corresponding to the student's family information include: at least one of the father's education level, the mother's education level, the father's occupation, and the mother's occupation; the variables corresponding to the student's school status include: at least one of health status, school mode, and attendance; and the variables corresponding to the student's character include: personality, way of collecting information, way of thinking, way of execution, belief, emotional stability, interest, self-efficacy, classroom concentration, divergent thinking ability, and learning communication ability.

10. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1 to 6.

Images