Digital teaching evaluation system based on machine learning

Abstract

The invention discloses a digital teaching evaluation system based on machine learning, and the system comprises a data collection module which is used for collecting multi-dimensional teaching related data, and the multi-dimensional teaching related data comprises multi-dimensional learning behavior data of students, multi-dimensional teaching behavior data of teachers, and external education resource data; and the data preprocessing module is connected with the data acquisition module and is used for cleaning and standardizing the acquired data, and in the data preprocessing module, when missing values are processed, numerical data are estimated and filled according to a mean value or data in the same group, and the data in the data preprocessing module are processed. And adopting a processing mode of deleting or marking the text type data as non-valid data according to the missing proportion. The model building module selects a proper machine learning algorithm according to data characteristics and evaluation targets, and optimizes model performance by adjusting model parameters and a cross validation method, so that teaching can be evaluated more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of digital teaching evaluation, and in particular to a digital teaching evaluation system based on machine learning. Background Art

[0002] In the modern education system, teaching evaluation is a key link to improve teaching quality and promote educational development. By evaluating the teaching process and teaching effect, it can provide feedback to educators so that they can adjust their teaching strategies; at the same time, it can also let students understand their learning status and motivate them to improve their learning methods. With the rapid development of information technology, digital teaching evaluation system has gradually become a research hotspot in the field of education.

[0003] After searching the relevant technologies, it is found that the existing digital teaching evaluation technologies mainly have the following problems:

[0004] Existing teaching evaluation systems often rely on limited data sources, such as traditional test scores, coursework completion, etc. For example, the evaluation systems used by some schools only evaluate students' learning outcomes based on their final exam scores and regular homework scores, and evaluate teachers' teaching quality based on the average scores of the classes they teach. This single data source cannot fully reflect the students' learning process and the overall picture of teachers' teaching. There is a lack of consideration for students' learning behaviors outside the classroom, such as self-study and the use of online learning resources. Similarly, data on teachers' teaching innovations and extracurricular interactions with students are not fully included in the evaluation system.

[0005] The evaluation models used by existing teaching evaluation systems are often relatively simple and crude. For example, some systems use simple linear regression models or rule-based evaluation models. These models are difficult to capture the complex nonlinear relationship between the teaching process and learning effects. In the construction of evaluation models, there is a lack of attention to feature engineering, and features that are highly relevant to the evaluation objectives are not carefully selected and extracted, resulting in limited prediction capabilities of the models. Summary of the invention

[0006] The purpose of the present invention is to provide a digital teaching evaluation system based on machine learning to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A digital teaching evaluation system based on machine learning, including

[0009] A data collection module is used to collect multi-dimensional teaching-related data, wherein the multi-dimensional teaching-related data includes multi-dimensional learning behavior data of students, multi-dimensional teaching behavior data of teachers, and external educational resource data;

[0010] A data preprocessing module is connected to the data acquisition module and is used to clean and standardize the collected data. In the data preprocessing module, when processing missing values, numerical data is estimated and filled according to the mean or the same group of data, and text data is deleted or marked as having no valid data according to the missing ratio;

[0011] A feature engineering module, connected to the data preprocessing module, for performing feature extraction and feature selection on the preprocessed data;

[0012] A model building module, connected to the feature engineering module, for building and training a machine learning evaluation model;

[0013] A visualization module, connected to the model building module, is used to design a visualization interface according to different user needs and provide interactive functions;

[0014] The authority management module is connected to the visualization module and is used to classify different user roles and define authorities.

[0015] In the present invention, in the data collection module, the learning trajectory of students on the online learning platform includes learning time, learning frequency, and learning period distribution; the interaction data between students and learning resources includes the number of clicks, dwell time, and resource downloads; the activity of teachers in the teaching community includes sharing teaching experience and participating in discussions on teaching methods; the interaction data between teachers and students outside the classroom includes the number of email exchanges and the timeliness of answering questions.

[0016] In the present invention, the students' multi-dimensional learning behavior data include the students' learning trajectories on the online learning platform, the students' interaction data with learning resources, the students' discussion data in learning communities or forums outside the classroom, and the students' use of learning devices; the teachers' multi-dimensional teaching behavior data include the teachers' teaching resource update frequency, the teachers' activity in the teaching community, and the teachers' interaction data with students outside the classroom.

[0017] In the present invention, the data cleaning includes removing duplicate records and processing missing values, and data standardization uses the Z-score standardization method to standardize numerical data to a specific interval.

[0018] In the present invention, the feature extraction includes extracting learning concentration features and resource preference features from students' learning behavior data, and extracting teaching innovation tendency features and student care features from teachers' teaching behavior data. The feature selection uses correlation coefficient analysis and information gain method to remove features with low correlation and low contribution to classification.

[0019] In the present invention, the machine learning evaluation model includes using a support vector machine algorithm to construct a student learning effect evaluation model, with student-related feature data as input and student comprehensive learning outcomes as output, using a random forest algorithm to construct a teacher teaching effect evaluation model, with teacher-related feature data as input and teacher teaching quality evaluation level as output, and optimizing model performance by adjusting model parameters and cross-validation methods;

[0020] Suppose the input student-related feature data is x i =(x i1 ,x i2 ,...,x id ), i = 1, 2, ..., n, d is the feature dimension, n is the number of samples, and the corresponding student comprehensive learning outcome is y i , the objective function is:

[0021]

[0022] The constraints are:

[0023]

[0024] Among them, w represents the normal vector of the hyperplane, which determines the direction of the hyperplane, b represents the intercept of the hyperplane, and ξ i Represents a slack variable, which is used to deal with sample points that do not satisfy |yi-(w·xi+b)|≤∈. C represents the penalty coefficient, which is a positive hyperparameter used to weigh the importance of empirical risk and regularization terms. A larger C value indicates a higher requirement for the fitting of training samples, which may lead to overfitting. A smaller C value pays more attention to the generalization ability of the model. ∈ represents a parameter in the insensitive loss function, which defines a tube around the predicted value. As long as the error between the predicted value and the true value is within this tube, it is not counted in the loss.

[0025] Assume that the teacher-related feature data is x, and the corresponding teaching quality evaluation level is y. For a node t, the sample set is D t , the number of categories is K, and the formula of the Gini index is:

[0026]

[0027] Among them, p k is the proportion of samples belonging to the kth class in node t;

[0028] When a feature j and a split point s are selected to split node t into two child nodes D tL and D tR When , the Gini index after splitting is:

[0029]

[0030] During the construction process, the decision tree selects the feature j and split point s that maximize ΔGini for splitting.

[0031] In the present invention, in the model building module, the support vector machine model optimizes performance by adjusting the kernel function and the penalty coefficient C, and the random forest model optimizes performance by adjusting the number of decision trees and the depth of the trees.

[0032] In the present invention, the visualization module provides different visualization views for school administrators, teachers, and students, and the visualization interface includes but is not limited to zooming, filtering, and sorting functions.

[0033] In the present invention, the authority management module uses a role-based access control mechanism to verify the user identity and grant corresponding permissions. The user roles include school administrators, teachers, students and parents, and different roles have different viewing and operating permissions.

[0034] In the present invention, in the visualization module, the visualization view of the school administrator uses a bar chart or a radar chart to show the comparison of teaching effects of different grades and different majors, the visualization view of the teacher uses a line chart to show the curve of teaching effect changes over time and a box plot to show the comparison with teachers in the same group, and the visualization view of the student uses a pie chart to show the degree of knowledge mastery and a line chart to show the trend of learning ability improvement.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. The data collection module of the present invention collects multi-dimensional teaching-related data, which not only covers traditional teaching achievement data, but also includes students' multi-dimensional learning behavior data and teachers' multi-dimensional teaching behavior data, and also collects external educational resource data. Such a comprehensive data source can more completely reflect the teaching process and effect;

[0037] 2. The model building module of the present invention selects a suitable machine learning algorithm according to data characteristics and evaluation objectives, and optimizes model performance by adjusting model parameters and cross-validation methods, so that teaching can be evaluated more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a schematic diagram of the overall structure of a digital teaching evaluation system based on machine learning in the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] like Figure 1 As shown, the present invention provides a technical solution:

[0041] A digital teaching evaluation system based on machine learning, including

[0042] A data collection module is used to collect multi-dimensional teaching-related data, wherein the multi-dimensional teaching-related data includes multi-dimensional learning behavior data of students, multi-dimensional teaching behavior data of teachers, and external educational resource data;

[0043] In this embodiment, in the data collection module, the learning trajectory of students on the online learning platform includes learning time, learning frequency, and distribution of learning time periods; the interaction data between students and learning resources includes the number of clicks, dwell time, and resource downloads; the teacher's activity in the teaching community includes sharing teaching experience and participating in discussions on teaching methods; the interaction data between teachers and students outside the classroom includes the number of email exchanges and the timeliness of answering questions; the students' multi-dimensional learning behavior data include the students' learning trajectory on the online learning platform, the interaction data between students and learning resources, the discussion data of students participating in learning communities or forums outside the classroom, and the use of students' learning devices; the teachers' multi-dimensional teaching behavior data include the frequency of teachers' teaching resource updates, the teachers' activity in the teaching community, and the interaction data between teachers and students outside the classroom.

[0044] Specifically, this module aims to comprehensively collect various types of data related to teaching evaluation. For students' multi-dimensional learning behavior data, learning trajectory information is obtained by embedding data collection tools in the online learning platform, such as recording the time when students log in to the platform, the duration of each learning session, the frequency of learning, and the distribution of learning in different time periods (such as daytime and nighttime). Learning trajectory data can reflect students' learning habits and degree of involvement. When students interact with learning resources, the platform will record interactive data such as the number of clicks, the time spent on different resource pages, and the resource download status. These data are helpful in analyzing students' preference for different types of learning resources (such as videos, documents, exercises, etc.). In terms of learning communities or forums, by monitoring students' posting and replying behaviors, counting the number of questions and analyzing the quality of answers, we can understand students' active learning and knowledge exchange outside the classroom. At the same time, we can obtain students' learning device usage, such as whether they use mobile devices for learning, the performance parameters of the devices, etc. The device type and performance may affect the learning experience and effect. For teachers' multi-dimensional teaching behavior data, the system will regularly check the teacher's teaching resource library and count the frequency of teaching resource updates, which reflects the teacher's enthusiasm for updating teaching content. Monitor teachers' activities in the teaching community, such as sharing teaching experience and participating in discussions on teaching methods, to measure teachers' awareness of teaching innovation and teamwork ability. Evaluate teachers' attention to students by analyzing interactive data such as the number of email exchanges between teachers and students outside the classroom and the timeliness of answering questions. In addition, collect external educational resource data and obtain the latest educational research results in the industry, the dynamics of vocational skills demand in related fields, and other information through web crawler technology. These external data can provide a reference for evaluating the matching degree between teaching content and actual needs.

[0045] A data preprocessing module is connected to the data acquisition module and is used to clean and standardize the collected data. In the data preprocessing module, when processing missing values, numerical data is estimated and filled according to the mean or the same group of data, and text data is deleted or marked as having no valid data according to the missing ratio;

[0046] In this embodiment, the data cleaning includes removing duplicate records and processing missing values, and data standardization uses the Z-score standardization method to standardize numerical data to a specific interval;

[0047] Specifically, the data collected by the data acquisition module will be transmitted to the data preprocessing module. In the process of data cleaning, for the processing of duplicate records, such as multiple records of the same learning behavior that may be generated by students on the learning platform due to network fluctuations, the system will identify and remove duplicates by comparing the key identifiers of the data (such as timestamps, operation types, etc.) to ensure the accuracy of the data. In terms of processing missing values, for numerical data, such as missing values ​​of students' learning time, if the overall data distribution is relatively regular, the average learning time of students can be calculated to fill the missing part; if the data has grouping characteristics (such as grouping by grade or class), it can be estimated and filled based on the average learning time of students in the same group. For text data, such as students' discussion data in the learning community, if the missing ratio is small (such as less than 10%), the corresponding records are directly deleted, because a small number of missing records have little impact on the overall analysis; if the missing ratio is large (such as more than 30%), it is marked as "no valid data" so that it can be identified and specially processed in subsequent analysis. Data standardization uses the Z-score standardization method. Taking the student's test scores as an example, first calculate the mean μ and standard deviation σ of all students' test scores, and then for each student's score x, according to the formula Calculate and convert the scores to the interval [0,1]. The purpose of this is to eliminate the differences in dimensions between different data. For example, the numerical range and meaning of students' study time and test scores are different. Standardization makes it easier to perform data analysis and model construction on the same scale.

[0048] A feature engineering module, connected to the data preprocessing module, for performing feature extraction and feature selection on the preprocessed data;

[0049] In this embodiment, the feature extraction includes extracting learning concentration features and resource preference features from students' learning behavior data, and extracting teaching innovation tendency features and student care features from teachers' teaching behavior data. The feature selection uses correlation coefficient analysis and information gain method to remove features with low correlation and low contribution to classification.

[0050] Specifically, clean and standardized data after data preprocessing enters the feature engineering module. In the feature extraction process, the learning concentration feature is calculated from the student's learning trajectory data. For example, by analyzing the distribution of learning time periods, if the student's learning frequency is high and the learning time is stable in a concentrated time period, it can be considered that his learning concentration is high; conversely, if the learning time period is scattered and the duration fluctuates greatly, the concentration is low. From the interaction data between students and learning resources, the resource preference feature is determined according to the proportional relationship between the number of clicks and the dwell time. For example, if the number of clicks on video resources is large and the dwell time is long, it indicates a high preference for video resources. For teachers, the teaching innovation tendency feature is extracted from the teaching resource update frequency and the teaching community activity. If the teacher frequently updates the teaching resources and actively participates in the discussion of innovative methods in the teaching community, then his teaching innovation tendency is high. From the interaction data between teachers and students outside the classroom, the teacher's student care feature is measured according to the depth of email communication and the timeliness of answering questions. Teachers who answer students' questions in a timely and in-depth manner have a higher degree of care. In the feature selection stage, the correlation coefficient analysis method is used. Taking the students' knowledge mastery (which can be measured by test scores, etc.) as the target, calculate the correlation coefficient between each extracted feature and the knowledge mastery. For example, if the correlation coefficient between a feature and the knowledge mastery is close to 0, it means that the correlation between the feature and the target is low. At the same time, the information gain method is used to evaluate the contribution of each feature to the classification (such as classifying students' learning effects into excellent, good, qualified, and unqualified grades). Assuming that there is a feature whose distribution difference under different learning effect levels is very small, then the information gain of this feature is low. In this way, the most valuable features for teaching evaluation are selected, redundant and irrelevant features are removed, the model calculation amount is reduced, and the model accuracy is improved.

[0051] A model building module, connected to the feature engineering module, for building and training a machine learning evaluation model;

[0052] In this embodiment, in the model construction module, the support vector machine model optimizes performance by adjusting the kernel function and the penalty coefficient C, and the random forest model optimizes performance by adjusting the number of decision trees and the depth of the tree. The machine learning evaluation model includes using the support vector machine algorithm to construct a student learning effect evaluation model, with student-related feature data as input and student comprehensive learning outcomes as output, and using the random forest algorithm to construct a teacher teaching effect evaluation model, with teacher-related feature data as input and teacher teaching quality evaluation level as output, and optimizing model performance by adjusting model parameters and cross-validation methods;

[0053] Suppose the input student-related feature data is x i =(x i1 ,x i2 ,...,x id), i = 1, 2, ..., n, d is the feature dimension, n is the number of samples, and the corresponding student comprehensive learning outcome is y i , the objective function is:

[0054]

[0055] The constraints are:

[0056]

[0057] Among them, w represents the normal vector of the hyperplane, which determines the direction of the hyperplane, b represents the intercept of the hyperplane, and ξ i Represents a slack variable, which is used to deal with sample points that do not satisfy |yi-(w·xi+b)|≤∈. C represents the penalty coefficient, which is a positive hyperparameter used to weigh the importance of empirical risk and regularization terms. A larger C value indicates a higher requirement for the fitting of training samples, which may lead to overfitting. A smaller C value pays more attention to the generalization ability of the model. ∈ represents a parameter in the insensitive loss function, which defines a tube around the predicted value. As long as the error between the predicted value and the true value is within this tube, it is not counted in the loss.

[0058] In a specific embodiment, suppose there is an online learning platform and 100 students participate in multiple courses. The system collects multi-dimensional learning behavior data such as students' learning time, learning frequency, participation in the learning community, and external educational resource data such as course difficulty coefficient and course resource type. These data will be used as input features, and the final grades of students in these courses (comprehensive learning outcomes) will be used as output; a hyperplane y = w·x+b is used to fit the student's feature data x and comprehensive learning outcomes y. For each sample point (x i ,y i ), define a loss function; in support vector regression, the ∈-insensitive loss function is used. For the sample point (x i ,y i ), if the predicted value and the true value y i The error is within the range of ∈, that is, |yi-(w·xi+b)|≤∈, indicating that there is no loss for this sample point; but if |yi-(w·xi+b)|>∈, a loss occurs, and the slack variable ξ is introduced. i and Make y i -(w·x i +b)≤∈+ξ i and (w·x i +b)-y i ≤∈+ξ i , the total loss function is Among them, C is the penalty coefficient, which is used to weigh the empirical risk (fitting error) and model complexity (regularization term); in order to prevent the model from overfitting, the regularization term is added The final objective function and constraints are shown above;

[0059] Specifically, collect and preprocess students' multi-dimensional learning behavior data and external educational resource data, and organize them into feature matrix x and corresponding target vector y. Initialize the hyperplane parameters w and b, as well as the slack variables ξ and ξ * The quadratic programming algorithm is used to solve the above objective function. The quadratic programming algorithm will find an optimal set of w, b, ξ and ξ * , so that the objective function is minimized. In the solution process, the optimal hyperplane is found by continuously adjusting the values ​​of w and b. For example, in each iteration, the values ​​of w and b are updated according to the current gradient information (partial derivatives of the objective function with respect to w and b), moving in the direction of reducing the objective function. For the new student feature data x new , using the trained model y new =w·x new +b to predict its comprehensive learning outcomes.

[0060] Assume that the teacher-related feature data is x, and the corresponding teaching quality evaluation level is y. For a node t, the sample set is D t , the number of categories is K, and the formula of the Gini index is:

[0061]

[0062] Among them, p k is the proportion of samples belonging to the kth class in node t;

[0063] When a feature j and a split point s are selected to split node t into two child nodes D tL and D tR When , the Gini index after splitting is:

[0064]

[0065] During the construction process of the decision tree, the feature j and split point s that maximize ΔGini are selected for splitting;

[0066] In a specific embodiment, suppose there is a school with 50 teachers. The system collects multi-dimensional teaching behavior data such as the frequency of teachers' teaching resource updates, their activity in the teaching community, and their interaction data with students outside the classroom. The teachers' teaching quality evaluation levels are divided into four levels: excellent, good, qualified, and unqualified. These data will be used as input features and the evaluation level will be used as output. Suppose that node t has N samples, and these samples belong to K different teaching quality evaluation levels. Assume that K = 4 in the scenario; for the kth evaluation level, let the number of samples belonging to this level in node t be n k , then the proportion of this level in node t Gini Index For example, there are 10 samples in node t, of which 3 are excellent, 3 are good, 2 are qualified, and 2 are unqualified. When considering splitting node t into two child nodes D tL and D tR When D tL YesN L samples, D tR YesN R samples, and N = N L +N R ; For child node D tL and D tR , calculate their Gini index Gini (D tL ) and Gini(D tR ), the Gini index after splitting is as shown above, and the splitting method with the largest ΔGini is selected.

[0067] A visualization module, connected to the model building module, is used to design a visualization interface according to different user needs and provide interactive functions;

[0068] In this embodiment, the visualization module provides different visualization views for school administrators, teachers, and students, and the visualization interface includes but is not limited to zooming, filtering, and sorting functions. In the visualization module, the visualization view of school administrators uses a bar chart or a radar chart to show the comparison of teaching effects of different grades and different majors, the visualization view of teachers uses a line chart to show the curve of teaching effect changes over time and a box plot to show the comparison with teachers in the same group, and the visualization view of students uses a pie chart to show the degree of knowledge mastery and a line chart to show the trend of learning ability improvement;

[0069] Specifically, the evaluation results output by the model building module are passed to the visualization module. For school managers, since they need to grasp the overall teaching situation of the school from a macro perspective, the system uses bar charts or radar charts for visualization based on the evaluation results of students' learning effects and teachers' teaching effects of different grades and majors. For example, a bar chart can intuitively compare the differences in the average academic performance of students or the teaching quality evaluation levels of teachers between different grades or majors; a radar chart can show the comprehensive performance of multiple evaluation indicators (such as students' knowledge mastery, skill improvement, teachers' teaching innovation tendency, student care, etc.) between different grades or majors. For teachers, the system uses a line chart to display the changes in their personal teaching effect evaluation results over time. Teachers can clearly see the fluctuation trend of their teaching effects in different teaching cycles (such as semesters and school years) so as to adjust their teaching methods in a timely manner. At the same time, the box and whisker chart shows the comparison with the same group of teachers on various teaching evaluation indicators. The box and whisker chart can intuitively display the quartiles, outliers and other information of the data, so that teachers can understand their position and advantages and disadvantages in the team. For students, based on their personal learning effect evaluation results, pie charts are used to show the proportional relationship of knowledge mastery, such as the mastery ratio of knowledge in different subjects. Line charts are used to show the trend of learning ability improvement, such as the improvement of students' learning concentration and resource utilization efficiency over time. The visualization interface also provides interactive functions. For example, school administrators can use the filtering function to select specific grades or majors in bar charts or radar charts to view detailed data; teachers can use the sorting function to sort and view teachers in the same group according to different evaluation indicators in the box chart; students can use the zoom function to view the specific changes in their learning ability improvement in a certain period of time in the line chart.

[0070] A rights management module, connected to the visualization module, for classifying different user roles and defining rights;

[0071] In this embodiment, the authority management module uses a role-based access control mechanism to verify the user identity and grant corresponding permissions. The user roles include school administrators, teachers, students and parents, and different roles have different viewing and operation permissions.

[0072] Specifically, when a user logs in to the digital teaching evaluation system based on machine learning, the authority management module first obtains the user's login information and identifies the user's identity. According to the pre-defined user role classification (school administrators, teachers, students and parents), the user is classified into the corresponding role. For school administrators, since they need to comprehensively manage the teaching evaluation work, the system grants them the highest authority. This means that school administrators can view the teaching evaluation data of the entire school, including the students' learning effect evaluation results, the teachers' teaching effect evaluation results, the operating parameters of each module, etc. In addition, school administrators have the right to modify the evaluation indicators, such as adjusting the weights of the knowledge mastery and skill improvement in the students' learning effect evaluation according to the school's educational goals; they can also adjust the model parameters in the model building module, such as changing the penalty coefficient C in the support vector machine model or the number of decision trees in the random forest model. After the teacher logs in to the system, the authority management module grants corresponding permissions according to his role. Teachers can only view their own teaching evaluation results and the learning statistics of the students they teach, such as the average academic performance of students, the distribution of learning concentration, etc. Teachers cannot modify the evaluation indicators, which is to ensure the objectivity and consistency of the evaluation system. After students log in, they can only view their own learning evaluation results, including visual reports on their knowledge mastery of various subjects, learning ability improvement trends, etc. After parents log in, the system grants them the authority to view their children's learning evaluation results and some teachers' basic teaching information (such as the teacher's name, subjects taught, etc.) according to their identities. The permission management module adopts a role-based access control (RBAC) mechanism. When a user performs any operation (such as viewing data, modifying parameters, etc.), it will verify in real time whether the user's operation is within the scope of their authority. If a user attempts to perform an operation beyond the scope of his or her authority, the system will reject the operation and prompt the user that the authority is insufficient.

[0073] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0074] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A digital teaching evaluation system based on machine learning, characterized by: include A data collection module is used to collect multi-dimensional teaching-related data, wherein the multi-dimensional teaching-related data includes students' multi-dimensional learning behavior data, teachers' multi-dimensional teaching behavior data and external educational resource data; A data preprocessing module is connected to the data acquisition module and is used to clean and standardize the collected data. In the data preprocessing module, when processing missing values, numerical data is estimated and filled according to the mean or the same group of data, and text data is deleted or marked as having no valid data according to the missing ratio; A feature engineering module, connected to the data preprocessing module, for performing feature extraction and feature selection on the preprocessed data; A model building module, connected to the feature engineering module, for building and training a machine learning evaluation model; A visualization module, connected to the model building module, is used to design a visualization interface according to different user needs and provide interactive functions; The authority management module is connected to the visualization module and is used to classify different user roles and define authorities.

2. A digital teaching evaluation system based on machine learning according to claim 1, characterized in that: In the data collection module, the learning trajectory of students on the online learning platform includes learning time, learning frequency, and distribution of learning periods; the interaction data between students and learning resources includes the number of clicks, dwell time, and resource downloads; the activity of teachers in the teaching community includes sharing teaching experience and participating in discussions on teaching methods; the interaction data between teachers and students outside the classroom includes the number of email exchanges and the timeliness of answering questions.

3. A digital teaching evaluation system based on machine learning according to claim 1, characterized in that: The students' multi-dimensional learning behavior data include the students' learning trajectories on the online learning platform, the students' interaction data with learning resources, the students' discussion data in learning communities or forums outside the classroom, and the students' usage of learning devices; the teachers' multi-dimensional teaching behavior data include the frequency of teachers' teaching resource updates, the teachers' activity in the teaching community, and the interaction data between teachers and students outside the classroom.

4. A digital teaching evaluation system based on machine learning according to claim 1, characterized in that: The data cleaning includes removing duplicate records and processing missing values, and data standardization uses the Z-score standardization method to standardize numerical data to a specific interval.

5. The digital teaching evaluation system based on machine learning according to claim 1, characterized in that: The feature extraction includes extracting learning concentration features and resource preference features from students' learning behavior data, and extracting teaching innovation tendency features and student care features from teachers' teaching behavior data. The feature selection uses correlation coefficient analysis and information gain method to remove features with low correlation and low contribution to classification.

6. The digital teaching evaluation system based on machine learning according to claim 1, characterized in that: The machine learning evaluation model includes using a support vector machine algorithm to construct a student learning effect evaluation model, taking student-related feature data as input and student comprehensive learning outcomes as output, using a random forest algorithm to construct a teacher teaching effect evaluation model, taking teacher-related feature data as input and teacher teaching quality evaluation level as output, and optimizing model performance by adjusting model parameters and cross-validation methods; Suppose the input student-related feature data is x i =(x i1 ,x i2 ,...,x id ), i = 1, 2, ..., n, d is the feature dimension, n is the number of samples, and the corresponding student comprehensive learning outcome is y i , the objective function is: The constraints are: Among them, w represents the normal vector of the hyperplane, which determines the direction of the hyperplane, b represents the intercept of the hyperplane, and ξ i Represents a slack variable, which is used to deal with sample points that do not satisfy |yi-(w·xi+b)|≤∈. C represents the penalty coefficient, which is a positive hyperparameter used to weigh the importance of empirical risk and regularization terms. A larger C value indicates a higher requirement for the fitting of training samples, which may lead to overfitting. A smaller C value pays more attention to the generalization ability of the model. ∈ represents a parameter in the insensitive loss function, which defines a tube around the predicted value. As long as the error between the predicted value and the true value is within this tube, it is not counted in the loss. Assume that the teacher-related feature data is x, and the corresponding teaching quality evaluation level is y. For a node t, the sample set is D t , the number of categories is K, and the formula of the Gini index is: Among them, p k is the proportion of samples belonging to the kth class in node t; When a feature j and a split point s are selected, node t is split into two child nodes D tL and D tR When , the Gini index after splitting is: During the construction process, the decision tree selects the feature j and split point s that maximize ΔGini for splitting.

7. The digital teaching evaluation system based on machine learning according to claim 1, characterized in that: In the model building module, the support vector machine model optimizes performance by adjusting the kernel function and the penalty coefficient C, and the random forest model optimizes performance by adjusting the number of decision trees and the depth of the trees.

8. The digital teaching evaluation system based on machine learning according to claim 1, characterized in that: The visualization module provides different visualization views for school administrators, teachers, and students, and the visualization interface includes but is not limited to zooming, filtering, and sorting functions.

9. The digital teaching evaluation system based on machine learning according to claim 1, characterized in that: The authority management module uses a role-based access control mechanism to verify user identities and grant corresponding permissions. The user roles include school administrators, teachers, students and parents, and different roles have different viewing and operation permissions.

10. The digital teaching evaluation system based on machine learning according to claim 1, characterized in that: In the visualization module, the visualization view of school administrators uses bar charts or radar charts to show the comparison of teaching effects in different grades and majors, the visualization view of teachers uses line charts to show the curve of teaching effect changes over time and box plots to show the comparison with teachers in the same group, and the visualization view of students uses pie charts to show the degree of knowledge mastery and line charts to show the trend of learning ability improvement.

Images