Education management knowledge tracking method and system based on multi-task learning

Abstract

The invention relates to the field of education management, in particular to an education management knowledge tracking method and system based on multi-task learning. An educational management knowledge tracking system based on multi-task learning comprises a learning data processing module, a knowledge tracking module and a learning resource recommendation module. According to the invention, by introducing a multi-task learning framework, deep analysis of student learning behavior data and answer data is realized, and comprehensive learning state evaluation is provided; meanwhile, deviation processing operation is introduced, the answering time and accuracy of students are analyzed, answer data deviation caused by guessing or random factors is eliminated, the reference performance of answer data is improved, and therefore prediction accuracy and efficiency are improved; the comprehensive evaluation and prediction capability enables the educator to more comprehensively understand the learning state of the student, and provides more targeted guidance and support for the student.

Description

Technical Field

[0001] The present invention relates to the field of education management, and in particular to an education management knowledge tracking method and system based on multi-task learning. Background Art

[0002] With the rapid development of educational informatization, the demand for personalized education and precise teaching is growing. In the traditional education model, teachers often need to rely on personal experience and intuition to evaluate students' learning status and knowledge mastery. This method is not only inefficient, but also lacks objectivity and accuracy. In order to solve this problem, the education field has begun to explore the use of data analysis and machine learning technology to track and manage students' learning process.

[0003] In the field of education management, knowledge tracing technology is widely used to assess students' mastery of specific knowledge points. Traditional knowledge tracing methods are mainly based on single-task models, such as using Bayesian knowledge tracing or item response theory models to predict students' mastery of a single knowledge point. However, these methods often ignore the diversity and complexity of students' learning behaviors, such as learning habits, answering speed, error patterns, etc. These factors are equally important for a comprehensive understanding of students' learning status.

[0004] In order to more comprehensively understand and predict students' learning status, multi-task learning technology has been introduced into the field of education management; MTL allows the model to learn multiple related tasks at the same time and share the knowledge representing the learning process, thereby improving the generalization ability and prediction accuracy of the model; although MTL has achieved remarkable results in other fields, it still faces challenges in the application of education management knowledge tracking, such as how to effectively integrate students' learning behavior data and answer data, how to design a model architecture that can simultaneously predict students' grades, learning behaviors and knowledge points, and how to provide students with personalized learning resource recommendations based on the prediction results; in response to the above challenges, the present invention proposes an education management knowledge tracking method and system based on multi-task learning, which provides a new technical support for education management and realizes accurate tracking and personalized guidance of students' learning process. Summary of the invention

[0005] The present invention, by introducing a multi-task learning framework, realizes in-depth analysis and comprehensive utilization of students' learning behavior data and answer data, predicts students' grades, learning behaviors and knowledge points, and provides a comprehensive learning status assessment; at the same time, it introduces deviation processing operations to analyze students' answering time and accuracy, eliminates answer data deviations caused by guesswork or random factors, and improves the reference value of answer data, thereby improving the accuracy and efficiency of prediction; this comprehensive assessment and prediction capability enables educators to understand students' learning status more comprehensively and provide students with more targeted guidance and support.

[0006] A knowledge tracking method for education management based on multi-task learning, including: For any student in the target student group, the learning behavior data and answer data of the student are obtained, and based on the answer time, answer result and knowledge points involved in each question in the student's answer data, the answer data is subjected to deviation processing to obtain the student's accurate answer data; Construct a multi-task learning model, input each student's learning behavior data and accurate answer data into the multi-task learning model, and output the student's predicted score, predicted learning behavior, and knowledge point mastery respectively; based on the obtained predicted learning behavior, judge the learning status of each student; At the same time, for any student, the student's knowledge point mastery status output by the multi-task learning model in the most recent n times is obtained, and the obtained n knowledge point mastery status is arranged in chronological order to obtain the student's learning resource recommendation data set; the learning resource recommendation data set is used to calculate the student's progress level and fluctuation level of each knowledge point, and combined with the most recent knowledge point mastery status, the recommended value of each knowledge point of the student is calculated, and the recommended value of the knowledge point indicates the importance of the current knowledge point to the student; the recommended value of each knowledge point obtained is used to provide learning resources for the student.

[0007] Preferably, the specific operation of performing deviation processing on the answer data is as follows: For any correct answer in the student's answer data, obtain the student's answer time t for the current question; according to the knowledge points involved in the current question, extract questions involving the same knowledge points from the answer data, and calculate the student's correct rate for the extracted questions involving the same knowledge points ; At the same time, calculate the correct rate of the current question for all students in the target student group ; Select the students who answered the current question correctly in the target student group, and calculate the average answer time of all selected students based on the answer time of the current question , and calculate the standard deviation of the answer time of all selected students on the current question ; Using the formula Calculate the answer time threshold , and set the accuracy threshold , if the student's accuracy Less than the accuracy threshold And the answer time t is less than the answer time threshold , then remove the current question from the student's answer data; finally obtain the student's accurate answer data.

[0008] Preferably, the recommended value of each knowledge point of the student is calculated, and the specific operation is as follows: For any student, based on the obtained learning resource recommendation dataset of the student, for any knowledge point, obtain the n mastery levels of the current knowledge point in chronological order , i=1,2,…,n; using the formula Calculate the progress level b of the current knowledge point; Indicates n levels of mastery The first level of mastery, Indicates n levels of mastery The last level of mastery; Calculate and obtain the n mastery levels of the current knowledge points Average mastery level , using the formula Calculate and obtain the fluctuation level c of the current knowledge point; Finally, using the formula Calculate and obtain the recommended value of the current knowledge point; , and is the weight coefficient, and > > .

[0009] Preferably, the multi-task learning model includes an input layer, a shared layer and a specific task layer; the shared layer includes an LSTM layer, a first fully connected layer and a feature fusion layer; the specific task layer includes a second fully connected layer, a first output layer, a first loss function layer, a third fully connected layer, a second output layer, a second loss function layer, a fourth fully connected layer, a third output layer and a third loss function layer.

[0010] Preferably, the input layer of the multi-task learning model is used to receive the student's learning behavior data and accurate answer data; The LSTM layer in the shared layer is used to extract the temporal features in the learning behavior data and the accurate answer data; the first fully connected layer is used to extract the non-temporal features in the learning behavior data and the accurate answer data; the feature fusion layer is used to fuse the temporal features and non-temporal features extracted by the temporal data processing unit and the non-temporal data processing unit to obtain a comprehensive feature vector; The second fully connected layer in the specific task layer is used to further extract features from the comprehensive feature vector output by the shared layer, and uses the ReLU activation function for nonlinear transformation; the first output layer is used to output the student's predicted score through a linear activation function; the first loss function layer is used to calculate the difference between the predicted score and the actual score using the mean square error loss function; the third fully connected layer is used to extract features related to learning behavior from the comprehensive feature vector; the second output layer is used to output the student's learning behavior prediction through the Softmax activation function, including multiple categories, each category corresponds to a probability value, indicating the possibility that the student is in that behavior category; the second loss function layer is used to measure the classification accuracy of the model using the cross entropy loss function; the fourth fully connected layer is used to extract features about the mastery of knowledge points in the comprehensive feature vector; the third output layer is used to output the predicted value of the mastery of knowledge points, indicating the student's mastery of a specific knowledge point; the third loss function layer is used to optimize the model using the mean square error loss function to measure the difference between the predicted mastery and the actual mastery.

[0011] Preferably, for the training of the multi-task learning model, the specific operations are as follows: Obtain learning behavior training data and question answering training data marked with corresponding task labels, and divide all learning behavior training data and question answering training data into a training set and a validation set; input the training set into the parameter-initialized multi-task learning model for training, and use the validation set to validate the multi-task learning model to obtain the validation result; determine whether the obtained validation result meets the training conditions, and if so, output the trained multi-task learning model; if not, continue to train the multi-task learning model through the training set; wherein, the task weights of the multi-task learning model are adjusted through an adaptive optimization algorithm.

[0012] Preferably, the task weights of the multi-task learning model are adjusted by an adaptive optimization algorithm, and the specific operations are as follows: At initialization, an initial weight is assigned to each task of the multi-task learning model , j=1, 2, 3; in each iteration, calculate the loss function of each task , and based on the loss function To evaluate the training status of each task; For any task, use the formula Calculate the task weight of the current task at the k+1th iteration ;in, represents the loss value of task j at the kth iteration; is a preset constant used to avoid division by zero errors; In any iteration, using the formula Calculate the total loss function of the current iteration ,in is the regularization coefficient, which is used to control the magnitude of task weight update; the total loss function Parameters for multi-task learning models Updates.

[0013] Preferably, an education management knowledge tracking system based on multi-task learning includes: The learning data processing module includes a learning behavior data acquisition unit, an answer data acquisition unit and a deviation processing unit; the learning behavior data acquisition module is used to acquire the learning behavior data of the students; the answer data acquisition module is used to acquire the answer data of the students; the deviation processing module is used to perform deviation processing operations on the answer data according to the answer time and answer results of each question in the acquired student answer data, so as to acquire the accurate answer data of the students; The knowledge tracking module is used to use each student's learning behavior data and accurate answer data as the input of the multi-task learning model, and output the student's predicted score, predicted learning behavior and knowledge point mastery respectively; The learning resource recommendation module is used to judge the learning status of each student based on the predicted learning behavior; at the same time, it obtains the student's knowledge point mastery status output by the multi-task learning model for the most recent several times, and sorts them by time to obtain the student's learning resource recommendation data set; the learning resource recommendation data set is used to calculate the student's progress level and fluctuation level of each knowledge point, and combined with the most recent knowledge point mastery status, the recommended value of each knowledge point of the student is calculated, and the recommended value of each knowledge point obtained is used to provide learning resources for students.

[0014] The present invention has the following advantages: 1. The present invention introduces a multi-task learning framework to achieve in-depth analysis and comprehensive utilization of students' learning behavior data and answer data, predict students' grades, learning behaviors and knowledge points, and provide a comprehensive learning status assessment; at the same time, it introduces deviation processing operations to analyze students' answering time and accuracy, eliminates answer data deviations caused by guesswork or random factors, and improves the reference value of answer data, thereby improving the accuracy and efficiency of predictions; this comprehensive assessment and prediction capability enables educators to understand students' learning status more comprehensively and provide students with more targeted guidance and support.

[0015] 2. The present invention obtains the students' mastery of knowledge points through a multi-task learning model, thereby identifying the students' weak links in which knowledge points, and calculates the recommended value of each knowledge point by analyzing the changes in the students' mastery of knowledge points in recent periods of time, rather than simply calculating by the mastery level; this recommendation mechanism based on the students' actual learning needs not only improves the relevance and effectiveness of learning resources, but also helps students allocate their learning time and energy more efficiently, and strengthens their learning in a targeted manner; in addition, this recommendation method also helps teachers and educational institutions optimize the allocation of teaching resources, ensuring that each student can obtain the educational resources that best suit their learning progress and ability level, thereby achieving truly personalized education. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the structure of the education management knowledge tracking system based on multi-task learning adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0018] Embodiment 1, a method for tracking education management knowledge based on multi-task learning, comprising: For any student in the target student group, the learning behavior data and answer data of the student are obtained. The learning behavior data includes the interaction between the student and the learning platform, such as login time, learning time, and page browsing history; the answer data includes the questions answered by the student, the answer time, the behavior during the answer process, etc.; these data can provide a basis for personalized teaching and learning resource recommendation; based on the answer time, answer results and knowledge points involved in each question in the student's answer data, the answer data is processed for deviation to obtain the student's accurate answer data, which more truly reflects the student's learning ability and knowledge point mastery; because when students answer questions, the following situations often occur, resulting in deviations in the analysis results, for example, students do not know a certain question, but can answer it by guessing or other methods. If the answer result is correct, in the subsequent data analysis, it may be considered that the student understands the question, but in fact the student does not understand it; accurate answer data is the key to effective knowledge tracking and learning status evaluation. They enable the model to more accurately predict students' learning outcomes and provide personalized learning suggestions; Construct a multi-task learning model that can handle multiple related tasks simultaneously, including predicting students' grades, learning behaviors, and knowledge points; input each student's learning behavior data and accurate answer data into the multi-task learning model, and output the student's predicted grades, predicted learning behaviors, and knowledge points. These prediction results are based on the patterns and associations learned by the model from the data, providing educators with in-depth insights into students' learning status and helping to develop personalized teaching plans and interventions; based on the predicted learning behavior obtained, judge each student's learning status, such as learning efficiency, learning habits, and possible difficulties. This judgment helps to identify students' learning problems in a timely manner, so that corresponding teaching strategies can be adopted, such as adjusting teaching content, providing additional tutoring, or changing teaching methods; At the same time, for any student, the student's knowledge point mastery status output by the multi-task learning model in the most recent n times is obtained, and the obtained n knowledge point mastery status is arranged in chronological order to obtain the student's learning resource recommendation data set; the learning resource recommendation data set is used to calculate the student's progress level and fluctuation level of each knowledge point, and combined with the most recent knowledge point mastery status, the recommended value of each knowledge point of the student is calculated. These recommended values ​​reflect the degree to which students need extra attention on specific knowledge points, and the obtained recommended values ​​are used to provide students with learning resources; personalized learning resource recommendations can improve learning efficiency and help students get more support where they need it, thereby accelerating learning progress and improving learning outcomes.

[0019] The specific operations for deviation processing of answer data are as follows: For any correct answer in the student's answer data, obtain the student's answer time t for the current question. The answer time can be used as an indicator to evaluate whether the student is likely to guess. If the answer time is abnormally short, it may indicate that the student has not thought enough and is guessing. Based on the knowledge points involved in the current question, extract questions involving the same knowledge points from the answer data, and calculate the student's correct rate for the extracted questions involving the same knowledge points. By analyzing students’ performance on the same knowledge point, we can evaluate whether students have really mastered the knowledge point or just happened to guess correctly on some questions; at the same time, we can calculate the accuracy rate of all students in the target student group on the current question. , in order to understand the difficulty of the question in the entire student group, which helps to identify the general difficulty level of the question and distinguish whether the student answered incorrectly because the question was difficult or because of insufficient knowledge; select the students who answered the current question correctly in the target student group, and calculate the average answering time of all selected students on the current question based on the answering time of all selected students on the current question , and calculate the standard deviation of the answer time of all selected students on the current question , the average answer time can reflect the general response speed of the student group to the questions, while the standard deviation can show the degree of variation of students' answer time, which helps to identify outliers; using the formula Calculate the answer time threshold , and set the accuracy threshold , if the student's accuracy Less than the accuracy threshold And the answer time t is less than the answer time threshold , then the current question will be removed from the student's answer data. Removing these answers that may be guessed can reduce the noise in the data, making subsequent analysis and model training more accurate, thereby improving the prediction accuracy of students' answering behavior; finally, the student's accurate answer data is obtained, which is crucial to building an effective knowledge tracking system because it relies on high-quality data to evaluate students' learning status and knowledge mastery.

[0020] Calculate the recommended value of each knowledge point of the student. The specific operations are as follows: For any student, based on the obtained learning resource recommendation dataset of the student, for any knowledge point, obtain the n mastery levels of the current knowledge point in chronological order , i=1,2,…,n; using the formula Calculate the progress level b of the current knowledge point; the progress level b reflects the speed and effect of the student's learning progress on a specific knowledge point; a higher progress level indicates that the student has made significant progress on the knowledge point, indicating that the student may be more interested in this knowledge or have a higher talent, so more additional resources can be provided for this knowledge point, so that the student can make greater progress; on the contrary, a lower progress level may indicate that the student has some difficulty in understanding the knowledge in this area or has reached a relatively sufficient state and it is difficult to continue to make progress. If the student spends too much time on this knowledge point, the final benefit will be very small, so the resource allocation of this knowledge point can be appropriately reduced; Indicates n levels of mastery The first level of mastery, Indicates n levels of mastery The last level of mastery; Calculate and obtain the n mastery levels of the current knowledge points Average mastery level , using the formula Calculate and obtain the fluctuation level c of the current knowledge point; the fluctuation level refers to the stability of the student's mastery of a certain knowledge point; it is measured by analyzing the changes in the student's mastery of the knowledge point over multiple assessments; the fluctuation level reflects whether the student's mastery of a specific knowledge point is stable; a higher fluctuation level may indicate that the student's understanding of the knowledge point is not stable enough and may require more practice and review. On the contrary, a lower fluctuation level indicates that the student's mastery of the knowledge point is relatively stable and does not require too much practice; Finally, using the formula Calculate and obtain the recommended value of the current knowledge point; , and is the weight coefficient, and > > ; By comprehensively considering the latest mastery, progress level and fluctuation level of knowledge points, the system can provide more personalized learning resource recommendations for each student, ensuring that resources are effectively allocated to where they are most needed.

[0021] The multi-task learning model includes an input layer, a shared layer and a specific task layer; the shared layer includes an LSTM layer, a first fully connected layer and a feature fusion layer; the specific task layer includes a second fully connected layer, a first output layer, a first loss function layer, a third fully connected layer, a second output layer, a second loss function layer, a fourth fully connected layer, a third output layer and a third loss function layer.

[0022] The input layer of the multi-task learning model is used to receive students’ learning behavior data and accurate answer data; The LSTM layer in the shared layer is used to extract the temporal features in the learning behavior data and the accurate answer data; the first fully connected layer is used to extract the non-temporal features in the learning behavior data and the accurate answer data; the feature fusion layer is used to fuse the temporal features and non-temporal features extracted by the temporal data processing unit and the non-temporal data processing unit to obtain a comprehensive feature vector; The second fully connected layer in the specific task layer is used to further extract features from the comprehensive feature vector output by the shared layer, and uses the ReLU activation function for nonlinear transformation; the first output layer is used to output the student's predicted score through a linear activation function; the first loss function layer is used to calculate the difference between the predicted score and the actual score using the mean square error loss function; the third fully connected layer is used to extract features related to learning behavior from the comprehensive feature vector; the second output layer is used to output the student's learning behavior prediction through the Softmax activation function, including multiple categories, each category corresponds to a probability value, indicating the possibility that the student is in that behavior category; the second loss function layer is used to measure the classification accuracy of the model using the cross entropy loss function; the fourth fully connected layer is used to extract features about the mastery of knowledge points in the comprehensive feature vector; the third output layer is used to output the predicted value of the mastery of knowledge points, indicating the student's mastery of a specific knowledge point; the third loss function layer is used to optimize the model using the mean square error loss function to measure the difference between the predicted mastery and the actual mastery.

[0023] For the training of the multi-task learning model, the specific operations are as follows: Obtain learning behavior training data and question answering training data marked with corresponding task labels, and divide all learning behavior training data and question answering training data into a training set and a validation set; input the training set into the parameter-initialized multi-task learning model for training, and use the validation set to validate the multi-task learning model to obtain the validation result; determine whether the obtained validation result meets the training conditions, and if so, output the trained multi-task learning model; if not, continue to train the multi-task learning model through the training set; wherein, the task weights of the multi-task learning model are adjusted through an adaptive optimization algorithm.

[0024] The task weights of the multi-task learning model are adjusted through an adaptive optimization algorithm. The specific operations are as follows: At initialization, an initial weight is assigned to each task of the multi-task learning model , j=1, 2, 3; in each iteration, calculate the loss function of each task , and based on the loss function To evaluate the training status of each task; For any task, use the formula Calculate the task weight of the current task at the k+1th iteration ;in, represents the loss value of task j at the kth iteration; is a preset constant used to avoid division by zero errors; In any iteration, using the formula Calculate the total loss function of the current iteration ,in is the regularization coefficient, which is used to control the magnitude of task weight update; the total loss function Parameters for multi-task learning models Updates.

[0025] Embodiment 2, an education management knowledge tracking system based on multi-task learning, such as Figure 1 As shown, including: The learning data processing module includes a learning behavior data acquisition unit, an answer data acquisition unit and a deviation processing unit; the learning behavior data acquisition module is used to acquire the learning behavior data of the students; the answer data acquisition module is used to acquire the answer data of the students; the deviation processing module is used to perform deviation processing operations on the answer data according to the answer time and answer results of each question in the acquired student answer data, so as to acquire the accurate answer data of the students; The knowledge tracking module is used to use each student's learning behavior data and accurate answer data as the input of the multi-task learning model, and output the student's predicted score, predicted learning behavior and knowledge point mastery respectively; The learning resource recommendation module is used to judge the learning status of each student based on the predicted learning behavior; at the same time, it obtains the student's knowledge point mastery status output by the multi-task learning model for the most recent several times, and sorts them by time to obtain the student's learning resource recommendation data set; the learning resource recommendation data set is used to calculate the student's progress level and fluctuation level of each knowledge point, and combined with the most recent knowledge point mastery status, the recommended value of each knowledge point of the student is calculated, and the recommended value of each knowledge point obtained is used to provide learning resources for students.

[0026] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for tracking educational management knowledge based on multi-task learning, characterized in that: include: For any student in the target student group, the learning behavior data and answer data of the student are obtained, and based on the answer time, answer result and knowledge points involved in each question in the student's answer data, the answer data is subjected to deviation processing to obtain the student's accurate answer data; Construct a multi-task learning model, input each student's learning behavior data and accurate answer data into the multi-task learning model, and output the student's predicted score, predicted learning behavior, and knowledge point mastery respectively; based on the obtained predicted learning behavior, judge the learning status of each student; At the same time, for any student, the student's knowledge point mastery status output by the multi-task learning model in the most recent n times is obtained, and the obtained n knowledge point mastery status is arranged in chronological order to obtain the student's learning resource recommendation data set; the learning resource recommendation data set is used to calculate the student's progress level and fluctuation level of each knowledge point, and combined with the most recent knowledge point mastery status, the recommended value of each knowledge point of the student is calculated, and the recommended value of the knowledge point indicates the importance of the current knowledge point to the student; the recommended value of each knowledge point obtained is used to provide learning resources for the student.

2. According to claim 1, a method for tracking educational management knowledge based on multi-task learning is characterized in that: The specific operations for deviation processing of answer data are as follows: For any correct answer in the student's answer data, obtain the student's answer time t for the current question; according to the knowledge points involved in the current question, extract questions involving the same knowledge points from the answer data, and calculate the student's correct rate for the extracted questions involving the same knowledge points ; At the same time, calculate the correct rate of the current question for all students in the target student group ; Select the students who answered the current question correctly in the target student group, and calculate the average answer time of all selected students based on the answer time of the current question. , and calculate the standard deviation of the answer time of all selected students on the current question ; Using the formula Calculate the answer time threshold , and set the accuracy threshold , if the student's accuracy Less than the accuracy threshold And the answer time t is less than the answer time threshold , then remove the current question from the student's answer data; finally obtain the student's accurate answer data.

3. According to claim 2, a method for tracking educational management knowledge based on multi-task learning is characterized in that: Calculate the recommended value of each knowledge point of the student. The specific operations are as follows: For any student, based on the obtained learning resource recommendation dataset of the student, for any knowledge point, obtain the n mastery levels of the current knowledge point in chronological order , i=1,2,…,n; Using the formula Calculate the progress level b of the current knowledge point; Indicates n levels of mastery The first level of mastery, Indicates n levels of mastery The last level of mastery; Calculate and obtain the n mastery levels of the current knowledge points Average mastery level , using the formula Calculate and obtain the fluctuation level c of the current knowledge point; Finally, using the formula Calculate and obtain the recommended value of the current knowledge point; , and is the weight coefficient, and > > .

4. According to claim 3, a method for tracking educational management knowledge based on multi-task learning is characterized in that: The multi-task learning model includes an input layer, a shared layer and a specific task layer; the shared layer includes an LSTM layer, a first fully connected layer and a feature fusion layer; the specific task layer includes a second fully connected layer, a first output layer, a first loss function layer, a third fully connected layer, a second output layer, a second loss function layer, a fourth fully connected layer, a third output layer and a third loss function layer.

5. The method for tracking educational management knowledge based on multi-task learning according to claim 4 is characterized in that: The input layer of the multi-task learning model is used to receive students’ learning behavior data and accurate answer data; The LSTM layer in the shared layer is used to extract temporal features from the learning behavior data and the accurate answer data; the first fully connected layer is used to extract non-temporal features from the learning behavior data and the accurate answer data; feature The fusion layer is used to fuse the temporal features and non-temporal features extracted by the temporal data processing unit and the non-temporal data processing unit to obtain a comprehensive feature vector; The second fully connected layer in the specific task layer is used to further extract features from the comprehensive feature vector output by the shared layer, and uses the ReLU activation function for nonlinear transformation; the first output layer is used to output the student's predicted score through a linear activation function; the first loss function layer is used to calculate the difference between the predicted score and the actual score using the mean square error loss function; the third fully connected layer is used to extract features related to learning behavior from the comprehensive feature vector; the second output layer is used to output the student's learning behavior prediction through the Softmax activation function, including multiple categories, each category corresponds to a probability value, indicating the possibility that the student is in that behavior category; the second loss function layer is used to measure the classification accuracy of the model using the cross entropy loss function; the fourth fully connected layer is used to extract features about the mastery of knowledge points in the comprehensive feature vector; the third output layer is used to output the predicted value of the mastery of knowledge points, indicating the student's mastery of a specific knowledge point; the third loss function layer is used to optimize the model using the mean square error loss function to measure the difference between the predicted mastery and the actual mastery.

6. The method for tracking educational management knowledge based on multi-task learning according to claim 5, characterized in that: For the training of the multi-task learning model, the specific operations are as follows: Obtain learning behavior training data and question answering training data marked with corresponding task labels, and divide all learning behavior training data and question answering training data into a training set and a validation set; input the training set into the parameter-initialized multi-task learning model for training, and use the validation set to validate the multi-task learning model to obtain the validation result; determine whether the obtained validation result meets the training conditions, and if so, output the trained multi-task learning model; if not, continue to train the multi-task learning model through the training set; wherein, the task weights of the multi-task learning model are adjusted through an adaptive optimization algorithm.

7. The method for tracking educational management knowledge based on multi-task learning according to claim 6, characterized in that: The task weights of the multi-task learning model are adjusted through an adaptive optimization algorithm. The specific operations are as follows: At initialization, an initial weight is assigned to each task of the multi-task learning model , j = 1, 2, 3; In each iteration, the loss function for each task is calculated , and based on the loss function To evaluate the training status of each task; For any task, use the formula Calculate the task weight of the current task at the k+1th iteration ;in, represents the loss value of task j at the kth iteration; is a preset constant used to avoid division by zero errors; In any iteration, using the formula Calculate the total loss function for the current iteration ,in is the regularization coefficient, which is used to control the magnitude of task weight update; the total loss function Parameters for multi-task learning models Updates.

8. An education management knowledge tracking system based on multi-task learning, characterized in that: The system is applied to the education management knowledge tracking method based on multi-task learning as described in any one of claims 1 to 7, comprising: The learning data processing module includes a learning behavior data acquisition unit, an answer data acquisition unit and a deviation processing unit; the learning behavior data acquisition module is used to acquire the learning behavior data of the students; the answer data acquisition module is used to acquire the answer data of the students; the deviation processing module is used to perform deviation processing operations on the answer data according to the answer time and answer results of each question in the acquired student answer data, so as to acquire the accurate answer data of the students; The knowledge tracking module is used to use each student's learning behavior data and accurate answer data as the input of the multi-task learning model, and output the student's predicted score, predicted learning behavior and knowledge point mastery respectively; The learning resource recommendation module is used to judge the learning status of each student based on the predicted learning behavior; at the same time, it obtains the student's knowledge point mastery status output by the multi-task learning model for the most recent several times, and sorts them by time to obtain the student's learning resource recommendation data set; the learning resource recommendation data set is used to calculate the student's progress level and fluctuation level of each knowledge point, and combined with the most recent knowledge point mastery status, the recommended value of each knowledge point of the student is calculated, and the recommended value of each knowledge point obtained is used to provide learning resources for students.

Images