Model construction method for identifying course planning quality influence factors based on machine learning

Abstract

The invention discloses a model construction method for identifying course planning quality influence factors based on machine learning, and the method comprises the steps: firstly, developing a course planning evaluation tool which covers four aspects of reasonability, consistency, regularity and feasibility, and is used for comprehensively evaluating the quality of a course planning scheme of a collected sample, thereby forming four types of labels; meanwhile, 25 features of the school are collected, and a label and feature data set is formed; secondly, a model network with a data input layer, a LassoNet layer and a multi-task learning layer is designed, relevant training is carried out, the four LassoNet models can identify key factors influencing the course planning quality in the four aspects of reasonability, consistency, regularity and feasibility in feature selection, it is ensured that the models pay attention to most relevant variables, and the accuracy of course planning is improved. Therefore, the accuracy of curriculum planning quality influence factor prediction is improved, training is carried out through a multi-task learning layer, and a model capable of identifying curriculum planning quality influence factors is constructed.

Description

Technical Field

[0001] The present invention belongs to the field of educational management technology, and in particular to a model building method for identifying factors affecting curriculum planning quality based on machine learning. Background Art

[0002] School curriculum planning is a series of curriculum-related decisions and arrangements that are essential to achieving educational goals and promoting student development. However, schools often find it difficult to gain clear and valuable insights into their curriculum "self-understanding" because those responsible for preparing curriculum plans have limited understanding of factors that affect the quality of school curriculum planning or use inappropriate research methods. Therefore, even if schools are committed to promoting student growth by enhancing their overall curriculum system, they are not sure how to effectively initiate improvements or updates to their existing curriculum plans. In the past, curriculum researchers have conducted curriculum planning research from several perspectives, including curriculum planning methods, curriculum planning frameworks, factors that affect curriculum planning, and curriculum content arrangements for specific professions or disciplines, but with little success.

[0003] With the emergence of advanced technologies such as artificial intelligence and machine learning, it is possible to analyze increasingly large and complex multidimensional data sets. However, since school curriculum planning is a personalized plan for the school, there is little data available for reference on the Internet, which limits the sample size of machine learning and makes it difficult to use machine learning and other methods to identify factors affecting the quality of curriculum planning. Summary of the invention

[0004] The present invention aims to solve the above-mentioned technical problems existing in the prior art and provides a model construction method for identifying factors affecting the quality of course planning based on machine learning.

[0005] The technical solution of the present invention is: a model construction method for identifying factors affecting the quality of curriculum planning based on machine learning, which is carried out according to the following steps:

[0006] Step 1. Collect school samples and preprocess them to form label data sets and feature data sets

[0007] Step 1.1 Collect school curriculum plans, use evaluation tools to score the rationality, consistency, regularity, and feasibility of curriculum plans, and determine four types of labels based on the scores to form a label data set;

[0008] Step 1.2 collects features related to the quality of school curriculum planning to form a feature data set;

[0009] Step 2. Build and train the network

[0010] Step 2.1 establishes a network, wherein the network has a data input layer, a LassoNet layer and a multi-task learning layer, wherein the LassoNet layer is composed of four parallel LassoNet models, and the multi-task learning layer is a multi-task learning model;

[0011] Step 2.2: Train the network as follows:

[0012] Step 2.2.1 Input the data of the feature dataset and label dataset into the LassoNet layer;

[0013] Step 2.2.2: Four parallel LassoNet models are used to train and predict the features that affect the rationality, consistency, regularity, and feasibility of curriculum planning. Based on the ROC scores of the trained and predicted features, the first M features with high ROC scores are selected as the key features that affect the quality of curriculum planning.

[0014] Step 2.2.3 summarizes all key features that affect the quality of the course, uses the features shared by at least two aspects as the input of the multi-task learning model, and conducts multi-task learning training in four aspects.

[0015] The step 1.1 is as follows:

[0016] The rationality, consistency, regularity and feasibility of the curriculum planning were scored on a 5-point scale, and each aspect included four contents;

[0017] The four elements of rationality are as follows: elaborating the school's educational philosophy, vision and mission; integrating the principles of national education policy and the concept of the new curriculum; clearly defining the qualities of graduates: development goals, educational goals; taking care of students' curriculum needs and society's expectations;

[0018] The four elements of consistency are as follows: the natural fit and unique attributes of the curriculum objectives and the school's educational philosophy; the consistency of curriculum design and support; the overall structure is consistent with the expected graduate image and national requirements; and it is internally consistent with national requirements;

[0019] The four elements of the regularity are as follows: the course objectives give priority to the individual development of students' strengths, potentials and interests; the course structure includes a comprehensive and balanced range of subjects; the course arrangement facilitates learning of course content and development of students' abilities; clear, specific and innovative implementation suggestions;

[0020] The four elements of feasibility are as follows: curriculum implementation includes multiple methods, including self-directed learning; specific implementation requirements for each course; strategic integration of resources, including human, financial, material, time, space, and technology; comprehensive and effective implementation of curriculum support measures, including organizational and teacher training;

[0021] Determine that each aspect with a score equal to or greater than 13 is labeled "1", otherwise it is labeled "0", forming a labeled dataset Y t ={0, 1}, t=1, 2, ... T, T=4 is the total number of tasks;

[0022] The step 1.2 specifically collects the following 25 features: school type, sponsor type, urban and rural type, region, school years, number of classes, floor area, building area, whether it is a popular school, number of students in school, number of students enrolled that year, number of graduates that year, honorary title of the principal, admission rate of ordinary schools, admission rate of famous schools, average score of the high school entrance examination that year, number of faculty and staff, number of full-time teachers, number of art teachers, number of young teachers, number of backbone teachers, number of teachers with master's degrees, number of senior teachers, proportion of young teachers and proportion of backbone teachers; to form the data set X t ∈R d , extract the features of sample k in evaluation task t, denoted as For each task t, the feature spaces corresponding to labels 0 and 1 are represented as and N represents the total number of samples, and d represents the total number of features.

[0023] The step 2.2.2 performs training predictions through the LassoNet model input to the output skip-layer residual connection, which enables the elements to have non-zero weights in the hidden units. The focus of training predictions is to minimize the empirical reconstruction error:

[0024]

[0025] subjectto||W j (1) || ∞ ≤M|θ j |, j = 1, 2, ..., d

[0026] in By sample The empirical loss of the training dataset composed of task t, W j (1) Represents the weight of feature j in the first hidden layer, and determines the feature subset with the highest score as the key features that affect the quality of course planning.

[0027] The multi-task learning model in step 2.2.3 adopts hard parameter sharing or soft parameter sharing to transfer knowledge. In the case of hard parameter sharing, the parameters in all hidden layers except the output layer are shared; the loss function L of the multi-task learning model with hard parameter sharing is MDN The weighted linear sum of the losses for each task is calculated as follows:

[0028]

[0029] Where: t represents the task weight parameter optimized using the gradient descent method, θ is the model parameter, and W is the weight parameter.

[0030] The present invention first develops a curriculum planning evaluation tool, which covers four aspects: rationality, consistency, regularity and feasibility. Each aspect is further subdivided into multiple evaluation criteria. A 5-point scoring system is used to make the evaluation results have a quantitative basis for comprehensively evaluating the quality of the curriculum planning scheme of the collected samples, thereby forming four types of labels. At the same time, the present invention collects 25 parameters including school type, sponsor type, urban and rural type, etc. as features to form a label and feature data set; secondly, a model network with a data input layer, a LassoNet layer and a multi-task learning layer is designed and relevant training is performed. The four LassoNet models can respectively identify the key factors that have the greatest impact on the quality of curriculum planning in terms of rationality, consistency, regularity and feasibility in feature selection, ensuring that the model focuses on the most relevant variables, thereby improving the accuracy of predicting the factors affecting the quality of curriculum planning, and then training through the multi-task learning layer to construct a model that can identify the factors affecting the quality of curriculum planning. Experiments show that the model has the advantages of high recognition accuracy and strong generalization ability. When each school formulates a curriculum plan, this model can be used to accurately identify the factors affecting the quality of curriculum planning, which is conducive to obtaining high-quality curriculum planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the network structure of the model of the embodiment of the present invention.

[0032] Figure 2 Schematic diagram of the importance analysis of key features output by the LassoNet layer in an embodiment of the present invention.

[0033] Figure 3 Schematic diagram of AUC for four tasks in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] A model construction method for identifying factors affecting curriculum planning quality based on machine learning of the present invention is carried out according to the following steps:

[0035] Step 1. Collect school samples and preprocess them to form label data sets and feature data sets

[0036] Step 1.1 Collect school curriculum plans, use evaluation tools to score the rationality, consistency, regularity, and feasibility of curriculum plans, and determine four types of labels based on the scores to form a label data set;

[0037] The evaluation tool of the embodiment of the present invention evaluates the curriculum plan based on four key aspects: rationality, consistency, regularity and feasibility. Each aspect consists of 4 evaluation criteria, a total of 16 evaluation criteria. Each criterion is evaluated using a 5-point scoring system, ranging from 1 point (completely non-compliant) to 5 points (completely compliant). The higher the score, the better the quality of the school's curriculum planning. The specific evaluation criteria are shown in Table 1.

[0038] Table 1

[0039]

[0040]

[0041] "Reasonableness" means that the school's curriculum planning should reflect a clear educational philosophy, including its vision, mission and principles. At the same time, the curriculum plan should be consistent with the requirements of national policies and the expectations of the community or region. In addition, a high-quality curriculum plan must clearly define the image or development goals of graduates so that all curriculum activities can work towards a unified goal.

[0042] "Coherence" evaluates whether there is a logical relationship between the different components of the curriculum plan. It mainly examines the consistency of course objectives with the school's educational philosophy, the consistency of course design and support, the consistency of the overall course structure with the expected outcomes of graduates, and the consistency of course design with national requirements.

[0043] "Regularity" is to evaluate whether the curriculum plan conforms to the law of student development. The present invention mainly considers the following aspects: assessing whether the curriculum objectives prioritize students' personal development, potential and interests; examining whether the curriculum covers a variety of subjects and maintains an appropriate balance between subjects; analyzing whether the curriculum arrangement is conducive to effective learning; and assessing whether the learning methods are constructive and encouraging teachers to adopt innovative teaching methods, such as independent, cooperative and inquiry-based learning.

[0044] Feasibility focuses primarily on assessing the potential for success of a curriculum plan and involves evaluating whether the curriculum plan is both specific and feasible. For example, it examines whether there are specific requirements to implement each curriculum and whether the necessary human, material, and financial resources are available to ensure the successful implementation of the curriculum plan.

[0045] The present invention selects 258 K12 schools as the sample P 1 -P 258, the evaluation of school curriculum planning for all samples was conducted by a panel of five curriculum experts who rated the curriculum plans individually using the evaluation tool. The average score given by the evaluators was then calculated to obtain the score for each dimension and the overall score for school curriculum planning. After the scoring was completed, the consistency of the ratings of the five experts was checked using the Kendall coefficient value. The fact that the Kendall coefficient value was 0.964 (P = 0.007 < 0.05) showed that the ratings of the experts were well coordinated, as the Kendall coefficient value was close to 1 and the significance level was less than 0.05 indicating a high degree of agreement between the experts. In addition, Cronbach values ​​were conducted to evaluate the reliability of the tool. The analysis showed an overall Cronbach of 0.964 and a Cronbach range of 0.952 to 0.978 for each dimension. The results showed that the evaluation tool had good reliability and internal consistency.

[0046] According to the evaluation results, the score of each aspect of the curriculum planning is determined to be equal to or greater than 13 points, and the label is "1", otherwise the label is "0", that is, the labels of the identification features "reasonable" and "irrational", "consistent" and "inconsistent", "regular" and "irregular", and "feasible" and "infeasible" are formed to form the label data set Y t ={0, 1}, t=1, 2, ... T, T=4 is the total number of tasks;

[0047] Step 1.2 Collect 25 features related to the quality of curriculum planning in 258 K12 schools x 1 -x 25 The specific characteristics are as follows: school type, sponsor type, urban and rural type, region, years of operation, number of classes, floor area, building area, whether it is a popular school, number of students in school, number of admissions in the year, number of graduates in the year, honorary title of the principal, admission rate of ordinary schools, admission rate of famous schools, average score of the high school entrance examination in the year, number of faculty and staff, number of full-time teachers, number of art teachers, number of young teachers, number of backbone teachers, number of teachers with master's degrees, number of senior teachers, proportion of young teachers and proportion of backbone teachers; form the data set X t ∈R d , extract the features of sample k in evaluation task t, denoted as For each task t, the feature spaces corresponding to labels 0 and 1 are represented as and N represents the total number of samples, and d represents the total number of features;

[0048] For any sample and The object of the present invention is to construct a function f:X t →Y t (t=1, 2, 3, 4), so that

[0049] Step 2. Build and train the network

[0050] Step 2.1 establishes a network, the network is as follows Figure 1 As shown: there are a data input layer, a LassoNet layer and a multi-task learning layer, the LassoNet layer is composed of four parallel LassoNet models, and the multi-task learning layer is a multi-task learning model;

[0051] Step 2.2: Train the network as follows:

[0052] Step 2.2.1 Input the data of the feature dataset and label dataset into the LassoNet layer;

[0053] Step 2.2.2: Four parallel LassoNet models are used to train and predict the features that affect the rationality, consistency, regularity, and feasibility of curriculum planning. Based on the ROC scores of the trained and predicted features, the first M features with high ROC scores are selected as the key features that affect the quality of curriculum planning.

[0054] Specifically, the training prediction is performed through the LassoNet model input to the output skip-layer residual connection, which enables the elements to have non-zero weights in the hidden units. The focus of the training prediction is to minimize the empirical reconstruction error:

[0055]

[0056] subjectto||W j (1) || ∞ ≤M|θ j |, j = 1, 2, ..., d

[0057] in By sample The empirical loss of the training dataset composed of task t, W j (1) represents the weight of feature j in the first hidden layer, and determines the feature subset with the highest score as the feature that affects the quality of course planning;

[0058] The importance scores (ROC values) of various features predicted by LassoNet model training are as follows Figure 2 As shown, the selected influencing factors that show the highest prediction performance are highlighted in red. Each dimension (aspect) contains unique features with significant impact. In addition, features shared with other dimensions (aspects) can be observed. The first M = 6 features with high ROC scores are selected as key features affecting the quality of course planning.

[0059] From the perspective of rationality, it is clear that all factors will have some impact on the prediction performance, but several factors have the most significant impact. Specifically, the average score of the high school entrance examination, the number of senior teachers, the number of students enrolled in the year, the proportion of young teachers, the building area, the number of teachers with a master's degree, and the region play a leading role in determining the results. Consistency, like rationality, is affected by all factors in addition to the type of sponsoring unit and whether it is a hot school. Among them, the six characteristics that have the greatest impact on the prediction results are the average score of the high school entrance examination, the number of teaching classes, the admission rate of ordinary schools, the years of operation, the proportion of young teachers, and the number of students enrolled in the year. Regularity is only affected by the six key characteristics of the average score of the high school entrance examination, the number of teaching classes, the number of senior teachers, the floor area, the years of operation, and the proportion of young teachers. There are 12 key characteristics that affect the feasibility prediction results, but not all characteristics have the same important impact. Among them, the average score of the high school entrance examination, the number of senior teachers, the admission rate of famous schools, the floor area, the years of operation, and the proportion of backbone teachers are the six most influential characteristics.

[0060] Step 2.2.3 summarizes all key features that affect the quality of the course, uses the features shared by at least two aspects as the input of the multi-task learning model, and conducts multi-task learning training in four aspects;

[0061] The multi-task learning model of the present invention adopts hard parameter sharing or soft parameter sharing to transfer knowledge, corresponding to T data sets {X t , Y t}, where t∈{1, 2, …, T-1, T}, the labels of these tasks can be represented as the complementary matrix C∈R N*T , in the case of hard parameter sharing, the parameters in all hidden layers except the output layer are shared; the loss function L of the multi-task learning model with hard parameter sharing is MDN The weighted linear sum of the losses for each task is calculated as follows:

[0062]

[0063] Where: t represents the task weight parameter optimized using the gradient descent method, θ is the model parameter, and W is the weight parameter.

[0064] The embodiment of the present invention summarizes all the key features that affect the quality of the course to obtain 12 key features, namely: the average score of the high school entrance examination in that year, the number of senior teachers, the proportion of young teachers, the years of operation, the number of students enrolled in that year, the number of teaching classes, the floor area, the building area, the region, the admission rate of ordinary schools, the admission rate of famous schools, and the proportion of backbone teachers. The features shared by at least two aspects (features that affect two or more aspects) are extracted, and the final 7 key features are obtained as the input of the multi-task learning model, and the multi-task learning training of four aspects is carried out. These seven key features are the most important factors for improving the quality of school curriculum planning, as shown in Table 3.

[0065] Table 3

[0066]

[0067] The average score of the high school entrance examination is the most critical factor affecting the school curriculum setting. It has an important impact on the rationality, consistency, regularity and feasibility of the school curriculum setting. The ROC value is between 700-1000, with the highest value of 907, and its regularity is particularly obvious. The "average score of the high school entrance examination" refers to the average score of students in different subjects after taking standardized tests. It represents the overall academic performance and proficiency of students in a school. In addition to the average score of the high school entrance examination, the number of students enrolled and the number of classes in the year are also factors directly related to students. These factors can all be included in the scope of students.

[0068] The number of senior teachers and the ratio of young teachers are two important factors second only to the average score. The number of senior teachers is an important factor affecting the rationality, regularity and feasibility of curriculum planning, and the ratio of young teachers is an important factor affecting the rationality, consistency and regularity of curriculum planning. The number of senior teachers and the ratio of young teachers are both directly related to teachers and can be classified as teachers.

[0069] The length of school operation is an important factor in ensuring the consistency, regularity and feasibility of curriculum planning. At the same time, the land area has a significant impact on regularity and feasibility. The length of school operation and the land area are listed as supporting conditions for school curriculum development and belong to the category of external resources.

[0070] The predictive capability of the indicators of the embodiments of the present invention:

[0071] AUC has recently been proposed as an alternative single-number metric for evaluating the predictive ability of learning algorithms. The AUC values ​​of the four tasks in the embodiment of the present invention are as follows: Figure 3 As shown. Figure 3 It can be seen that all four tasks achieved significant AUC values, with AUC values ​​ranging from 0.8030 to 0.9130, indicating that the accuracy of the present invention is high and has significantly superior superior prediction performance.

[0072] In summary, the model established by the present invention can be used by various schools to identify factors affecting the quality of curriculum planning, especially by determining 7 key features. Each school only needs to input the 7 key features into the model, and identify the features of "reasonable" and "irrational", "consistent" and "inconsistent", "regular" and "irregular", as well as "feasible" and "infeasible", which can effectively improve the quality of curriculum planning.

Claims

1. A model construction method for identifying factors affecting curriculum planning quality based on machine learning, characterized in that Follow these steps: Step 1. Collect school samples and preprocess them to form label data sets and feature data sets Step 1.1 Collect school curriculum plans, use evaluation tools to score the rationality, consistency, regularity, and feasibility of curriculum plans, and determine four types of labels based on the scores to form a label data set; Step 1.2 collects features related to the quality of school curriculum planning to form a feature data set; Step 2. Build and train the network Step 2.1 establishes a network, wherein the network has a data input layer, a LassoNet layer and a multi-task learning layer, wherein the LassoNet layer is composed of four parallel LassoNet models, and the multi-task learning layer is a multi-task learning model; Step 2.2: Train the network as follows: Step 2.2.1 Input the data of the feature dataset and label dataset into the LassoNet layer; Step 2.2.2: Four parallel LassoNet models are used to train and predict the features that affect the rationality, consistency, regularity, and feasibility of curriculum planning. Based on the ROC scores of the trained and predicted features, the first M features with high ROC scores are selected as the key features that affect the quality of curriculum planning. Step 2.2.3 summarizes all key features that affect the quality of the course, uses the features shared by at least two aspects as the input of the multi-task learning model, and conducts multi-task learning training in four aspects.

2. The model construction method for identifying factors affecting curriculum planning quality based on machine learning according to claim 1 is characterized in that The step 1.1 is specifically as follows: The rationality, consistency, regularity and feasibility of the curriculum planning were scored on a 5-point scale, and each aspect included four contents; The four elements of rationality are as follows: elaborating the school's educational philosophy, vision and mission; integrating the principles of national education policy and the concept of the new curriculum; clearly defining the qualities of graduates: development goals, educational goals; taking care of students' curriculum needs and society's expectations; The four elements of consistency are as follows: the natural fit and unique attributes of the curriculum objectives and the school's educational philosophy; the consistency of curriculum design and support; the overall structure is consistent with the expected graduate image and national requirements; and it is internally consistent with national requirements; The four elements of the regularity are as follows: the course objectives give priority to the individual development of students' strengths, potentials and interests; the course structure includes a comprehensive and balanced range of subjects; the course arrangement facilitates learning of course content and development of students' abilities; clear, specific and innovative implementation suggestions; The four elements of feasibility are as follows: curriculum implementation includes multiple methods, including self-directed learning; specific implementation requirements for each course; strategic integration of resources, including human, financial, material, time, space, and technology; comprehensive and effective implementation of curriculum support measures, including organizational and teacher training; Determine that the score of each aspect is equal to or greater than 13 points and the label is "1", otherwise the label is "0", forming a labeled data set Y t ={0, 1}, t=1, 2, ... T, T=4 is the total number of tasks.

3. The model construction method for identifying factors affecting curriculum planning quality based on machine learning according to claim 2 is characterized in that Step 1.2 specifically collects the following 25 features: school type, sponsor type, urban and rural type, region, years of operation, number of classes, floor area, building area, whether it is a popular school, number of students in school, number of admissions in the year, number of graduates in the year, honorary title of the principal, admission rate of ordinary schools, admission rate of famous schools, average score of the high school entrance examination in the year, number of faculty and staff, number of full-time teachers, number of art teachers, number of young teachers, number of backbone teachers, number of teachers with master's degrees, number of senior teachers, proportion of young teachers and proportion of backbone teachers; to form the data set X t ∈R d , extract the features of sample k in evaluation task t, denoted as For each task t, the feature spaces corresponding to labels 0 and 1 are represented as and N represents the total number of samples, and d represents the total number of features.

4. The model construction method for identifying factors affecting the quality of course planning based on machine learning according to claim 3 is characterized in that The step 2.2.2 performs training predictions through the LassoNet model input to the output skip-layer residual connection, which enables the elements to have non-zero weights in the hidden units. The focus of training predictions is to minimize the empirical reconstruction error: in By sample The empirical loss of the training dataset composed of task t, W j (1) Represents the weight of feature j in the first hidden layer, and determines the feature subset with the highest score as the key features that affect the quality of course planning.

5. The model building method for identifying factors affecting curriculum planning quality based on machine learning according to claim 4 is characterized in that: The multi-task learning model in step 2.2.3 adopts hard parameter sharing or soft parameter sharing to transfer knowledge for training. In the case of hard parameter sharing, the parameters in all hidden layers except the output layer are shared; the loss function L of the multi-task learning model with hard parameter sharing is MDN The weighted linear sum of the losses for each task is calculated as follows: Where: t represents the task weight parameter optimized using the gradient descent method, θ is the model parameter, and W is the weight parameter.

Images