Computational Thinking Training and Deep Learning Evaluation Model Construction Based on Scratch Modular Programming Course

To improve the algorithmic dimension, critical thinking, and problem-solving ability of computational thinking (CT) in students' programming courses, first, a programming teaching model is constructed based on the scratch modular programming course. Secondly, the design process of the teaching model and the problem-solving model of visual programming are studied. Finally, a deep learning (DL) evaluation model is constructed, and the effectiveness of the designed teaching model is analyzed and evaluated. The T-test result of paired samples of CT is t = −2.08, P < 0.05. There are significant differences in the results of the two tests, and the designed teaching model can cause changes in students' CT abilities. The results reveal that the effectiveness of the teaching model based on scratch modular programming has been verified on the basis of experiments. The post-test values of the dimensions of algorithmic thinking, critical thinking, collaborative thinking, and problem-solving thinking are all higher than the pretest values, and there are individual differences. The P values are all less than 0.05, which testifies that the CT training of the designed teaching model has the algorithm dimension, critical thinking, collaborative thinking, and problem-solving ability of students' CT. The post-test values of cognitive load are all lower than the pretest values, indicating that the model has a certain positive effect on reducing cognitive load, and there is a significant difference between the pretest and post-test. In the dimension of creative thinking, the P value is 0.218, and there is no obvious difference in the dimensions of creativity and self-efficacy. It can be found from the DL evaluation that the average value of the DL knowledge and skills dimensions is greater than 3.5, and college students can reach a certain standard level in terms of knowledge and skills. The mean value of the process and method dimensions is about 3.1, and the mean value of the emotional attitudes and values is 2.77. The process and method, as well as emotional attitude and values, need to be strengthened. The DL level of college students is relatively low, and it is necessary to improve their DL level from the perspective of knowledge and skills, processes and methods, emotional attitudes and values. This research makes up for the shortcomings of traditional programming and design software to a certain extent. It has a certain reference value for researchers and teachers to carry out programming teaching practice.


Introduction
Since the rise of artifcial intelligence (AI) as a national development strategy, AI education, programming education, and robot education have been highly valued in society, schools, and family education. Cultivating students' training thinking is an important aspect that needs attention in the programming course. Especially with the advent of the era of education informatization 2.0, all citizens are required to pay attention to the cultivation of computational thinking (CT) ability [1]. Te proposition of CT conforms to the trend of the information age and provides new experiences and methods for future curriculum construction and reform. CT, Empirical Tinking, and Logical Tinking are considered to be the three major thinking models in the scientifc thinking spectrum [2].
In actual programming teaching, teachers often focus on the results of problem-solving and ignore the changes in the entire thinking process of learners in solving and analyzing problems. Te works or assignments created by learners become the only yardstick for judging their CT level. Tis kind of training model cannot improve the thinking ability of students in a real sense [3]. Scratch programming can efectively improve students' CT ability. It can be used in many aspects of teaching, such as designing animations, designing games, and solving math problems [4]. Exploring the classroom teaching implementation strategy based on scratch graphical programming in line with the current curriculum has become the top priority of information technology teaching [5]. Jiang and Li (2021) designed a scratch course aimed at improving the CT of middle school students in the teaching practice research. Te result of the analysis indicated that scratch teaching improves the CT of middle school students to a certain extent [6]. Chung and Shamir (2020) studied the scratch graphical programming tool combined with the Machine Learning for Kids project to form an AI algorithm. It was applied to youth AI education for efect verifcation [7]. Riera et al. (2019) used the hardware logic of the perception based on the scratch teaching platform and control module as a supplement to the scratch language software logic. Tey also built an intelligent hardware learning system that integrates perception, processing, control, and virtual scenes [8]. Te research literature suggests that CT training is mainly focused on primary and secondary schools and high schools, and there are relatively few studies in higher education. Most of the students in college have not been exposed to programming before, so the research on the cultivation of CT in college students is particularly important [9]. How to efectively cultivate CT and innovate the teaching model of CT is an important issue for college educators. Te deep learning (DL) model advocates for learners to actively transfer their knowledge and apply it to solve complex problems in reality, which helps to improve learners' own critical thinking and the construction of new knowledge [10]. Te learning methods are diverse, focusing on the transfer and application of knowledge, and the DL evaluation model satisfes the active and comprehensive development of learners at various aspects and levels. It transforms the actual ability of programming into the training of programming thinking, starting from scientifc literacy and comprehensively inspecting students' knowledge and learning ability [11].
On account of the existing theoretical achievements, a teaching model of CT training for programming courses on the idea of scratch modular programming is proposed, which permeates the idea of modular programming in the whole process of learning. And it guides students to decompose complex problems into subproblems, realize the advantages of modular decomposition tasks, and compare CT results before and after. Moreover, the DL evaluation model is constructed to verify whether this model can effectively cultivate and train students' comprehensive analytical thinking abilities in programming teaching courses from diverse dimensions. Tis research has a certain positive infuence on improving the algorithmic dimension, critical thinking, and problem-solving ability of students' CT and reducing cognitive load. Te innovation of design lies in the combination of the scratch programming tool and CT training, and the use of the DL model for comprehensive evaluation of teaching design. Scratch programming tools can help students establish and train programming ideas, lay the foundation for learning a professional programming language, improve the transfer of students' information technology (IT) capabilities, and further make up for the defciencies of traditional programming and design software. It has vital guiding signifcance for teachers and researchers to perform programming teaching practice.  [12]. CT applies the conceptual principles of computer science to problem-solving, models the relevant aspects of the problem, and applies the most efective methods to solve the problem. Te CT process includes features such as problem structuring, data analysis, model building, algorithm design, solution implementation, and application migration [13,14].

Te Training Elements of CT in Programming Courses.
Te CT training model belongs to the subordinate concept of training mode. Figure 1 displays the specifc training elements.
Te elements in Figure 1 include training objectives, content, implementation process, and evaluation. Te training goal includes two aspects. One is the general goal orientation of the country, society, or school for the cultivation of students' CT, which is a macro-level goal requirement; the other is the curriculum goal of the CT training courses ofered. Te training content is mainly refected in the CT syllabus, teaching content, teaching principles, teaching management methods, etc. Evaluation is the fnal form of testing a practical achievement.  [16,17]. Te important performance of CT is refected in the instruction such as selection, loop, and condition in Scratch teaching from diferent perspectives [18]. Modular programming is a program design that divides a large program into several small program modules according to functions. Each small program module completes a certain function, and each module cooperates with each other to complete the program design of the entire function [19,20]. Te steps of modular programming are demonstrated in Figure 2.

Construction of a Teaching
As shown in Figure 2, the frst step is to analyze the problem and clarify the tasks to be solved. Te teacher guides the students to decompose and refne the task step by step and divides it into several subtasks. Each subtask only completes part of the complete function and can be implemented by functions. Next, it is necessary to determine the calling relationship between various modules, guide students to continuously debug, and optimize the calling relationship between modules. Finally, call and parallel processing are implemented in the main program.
Modular thinking training allows students to decompose complex problems into many small problems that are easy to solve. Tis type of thinking training can improve students' higher-order thinking abilities. Tis model is applied to the scratch classroom for teaching practice [21]. Te teaching process mainly includes fve basic links: creating a situation, refning knowledge, assigning tasks, practical operation, and summarizing and refecting [22]. Te structure of the designed teaching model based on the Scratch modular program is illustrated in Figure 3. Figure 3 signifes that in the teaching activities, it mainly includes designing the situation and guiding the theme, case demonstration and knowledge explanation, assigning tasks and guiding thinking, support and inspection and supervision, and summary and evaluation. Student activities mainly involve process design and software editing, problem decomposition, schema construction, motivation and refection, and feedback and improvement. In terms of CT goals, they mainly cover creativity, algorithmic thinking, collaboration, critical thinking, and problem-solving. In the process of teaching practice, in the frst step, the teacher uses the grouping function to divide the students into two groups according to the programming level. In the second step, the teacher guides the students to analyze the problem and uses the Scratch modular idea to demonstrate the operation to the students. In this process, students closely link old and new knowledge and reconstruct new cognition. In the third step, teachers ask challenging questions and students answer them. In the fourth step, teachers conduct inspections and supervision, answer questions, and students visualize the plan. Finally, under the guidance of teachers, students make summaries and conduct experimental refections and iterative improvements at this stage.

Visual Programming Problem-Solving Model.
Visual programming tools lead learners to contact the code language in the way of module splicing, which can make learners accept learning programming psychologically. Its main teaching function is to weaken the writing of programming code, emphasize the application of CT knowledge and methods, and enhance the learner's motivation [23]. Visual programming tools can describe and execute problems in real situations in a modular programming language according to the problemsolving plan [24]. Figure 4 reveals the visual programming problem-solving model.
As shown in Figure 4, a plan is formed through CT and methods, and a visual programming platform program is built according to the plan. After the platform is debugged, the solution to the problem is obtained and mapped to the real situation. Trough the real situation, questions can be raised and fed back to the CTmethods. Te real situation can further extract the elements in the real situation, such as people, things, things and rules. Finally, the problem is solved.  Computational Intelligence and Neuroscience 3

DL Concept Teory. DL, also known as Deep Structure
Learning, is the inherent law of learning sample data to automatically learn data features and complete tasks such as classifcation and regression. It has the ability to analyze, learn independently, and recognize data, such as text and images. Figure 5 presents the DL neural network model. From bottom to top are the input layer, hidden layer, and output layer. Features are extracted layer by layer according to the feature distribution of the underlying data [25,26].
Te DL neural network model in Figure 5 uses unsupervised learning from the input layer to the output layer. In other words, the training starts from the input layer and goes up layer by layer. Te parameters of each layer are trained layer by layer without calibration data. Tis training can be regarded as an unsupervised training process.
Assuming that x 1 and x 2 represent the sample features. Te input layer receives the sample data features and then outputs the evaluation result. Te input x can be expressed as the following equation: x Te following equation indicates the preactivation output z [l] i .
Te activation output of the hidden layer a [l] i can be written as follows: Here the superscript [l] represents the number of layers in the neural network, and the subscript stands for the number of neuron nodes. By training on large-scale data, representative feature information is obtained, thereby achieving the purpose of classifying and predicting sample data [27].   In the DL network model, the feature sample data X trained by CT, and the parameters w [1] , b [1] are input to the frst hidden layer, and z [1] is calculated, as indicated in the following equation:

Model Evaluation Calculation
When fnding out a [l] , the parameters w [2] , b [2] , and a [l] are input into the second hidden layer together to solve z [2] for the subsequent calculation of a [2] . Propagation continues according to such rules, and the whole process is called forward propagation [28]. Te calculation process is presented as the following equations: [2] , a [2] � g z [2] .

(5)
By analogy, z [l] , a [l] , and y can be expressed as the following equations: An error loss function is defned as the measurement standard to make the output feature training value y gradually approach the real value y. Te error loss function is defned as follows: Te neural network model is hoped to meet the probability under certain conditions, as shown in the following equation: (8) can be rewritten as (9).
Taking the logarithm of both sides of the above results, the simplifed result is shown in the following equation: Te larger the value of p(y | x), the smaller the loss. Te cross-entropy function is expressed as follows:

DL Features.
According to the DL model and related teaching research, the basic characteristics of DL are summarized. DL is specifcally summarized into six basic features [29], as expressed in Figure 6. First, DL emphasizes a high degree of engagement in learning. Students with strong learning motivation can actively discover the meaning of knowledge, communicate and cooperate with teachers and classmates, and strive to build a knowledge system to cultivate the ability to solve practical problems. Second, DL focuses on critical understanding. Students should pay attention to understanding learning and promote DL to occur. Tird, it emphasizes the integration and construction of knowledge to form a new knowledge structure. Fourth, DL attaches importance to knowledge transfer application and problem-solving. Fifth, DL pays attention to the overall development of the mind, Output layer Z Intput layer X ... Computational Intelligence and Neuroscience including comprehensive practical ability, operational ability, problem-solving ability, and innovative application ability. Sixth, DL emphasizes self-direction and lifelong characteristics [30].

DL Route.
A DL route is proposed based on DL theory, which is divided into 8 stages. First, the learning objectives and learning content are designed. Second, the learner's learning level is preassessed. Tird, a positive learning atmosphere is created to stimulate previous learning. Fourth, students acquire new knowledge. Fifth, they perform feature extraction on the acquired knowledge. Sixth, they carry out in-depth processing of knowledge. Ten, learners form new knowledge-cognition pairs. Finally, the learning efect is evaluated [31,32]. Figure 7 provides the specifc route.

Experimental
Steps. Tis work carried out a learning activity experiment based on the teaching model of Scratch modular programming. Figure 8 shows the specifc steps. First, the same teacher guided the basic knowledge of Scratch programming for a week. After learning, the students were asked to complete the after-class practice task "let the kitten move". In weeks 2 and 3, students completed the Scratch pretest questions and pretest questionnaires. Te pretest questionnaire included assessments of learners' CT skills, self-efcacy, and cognitive load. In the following nine weeks, teachers taught modular programming. In the last two weeks, post-test questionnaires and post-test questions were measured and data collected. Post-test questionnaires included the Computational Tinking Scale, the Self-Efcacy Scale, and the Cognitive Load Scale. Te DL model was used to evaluate the learning results.  Week 2 Week 3 Week 4-12 Week 14  Te pretest contains 10 multiple-choice questions (100%); the post-test includes 10 multiple-choice questions, each with 3 points (60%) and 10 judgment questions, each with 2 points (40%). Finally, they are converted to the percent system. Te pretest and post-test questions were adapted from CTt. CTS is a fve-point Likert scale consisting of 29 items. Te assessment tool mainly measures the level of learners' CT ability from fve dimensions: creativity, algorithmic thinking, collaboration, critical thinking, and problem-solving. Te validity and reliability of the scale were studied through exploratory factor analysis, confrmatory factor analysis, item variance analysis, internal consistency coefcient, and constancy analysis. Te analysis results show that the scale is an efective and reliable measurement tool for measuring students' CT ability. Tere are eight items in the Group Self-efcacy Questionnaire, which test individual judgments of group competence and assessments of group competence on upcoming tasks. Each item was scored on a fvepoint Likert scale, with 5 representing strongly agree and 1 representing strongly disagree.
Cognitive load measures include two dimensions: mental load and mental efort. Te Cronbach alpha values for the two dimensions are 0.92 and 0.84, respectively. In this study, the description of the scale was adjusted to a language suitable for students' understanding according to the teacher's specifc class content, and a pre-and post-test survey was conducted.
Te software uses SPSS25 version and Excel to implement the data analysis.

Test Environment for DL Evaluation.
Te parameter settings and operating environment of the DL evaluation model are exhibited in Table 1.

Paired-Sample T-Test for Student CT.
SPSS25 was used to analyze the overall changes of students' CT before and after the test. Te paired sample T test was used. Table 2 lists the results.
In Table 2, the mean of the pretest was 97.45, and the mean of the post-test was 99.28, showing a rise. Te standard deviation of the pretest was 13.38, and the standard deviation of the post-test was 12.42, t � -2.08, and p � 0.022 < 0.05. Te results of the two tests were signifcantly diferent, indicating that the teaching model based on modular programming can cause changes in students' CT ability.  Computational Intelligence and Neuroscience 7

Comparison of Dimensional Tests of Creativity and
Algorithmic Tinking. Te CT test was conducted on 10 students from the perspective of creative thinking and algorithmic thinking. Te results of the pretest and post-test were compared, as shown in Figure 9. Figure 9 denotes that the post-test scores of the 10 students' creative thinking are higher than the pretest. Student No. 3 has the highest score. Te pretest value is 4.3 the post-test score is 4.9. Student No. 2 has the lowest score, with a pretest value of 2.6 and a post-test value of 3.1. But overall, creative thinking is a growing trend. As displayed in Figure 9(b), the post-test values for the dimension of algorithmic thinking are all higher than the pretest values. Student No. 5 has the lowest score, with a pretest value of 3.7 and a post-test value of 4.1, and the overall trend is upward. Te post-test values are higher than the pretest values, and creativity and algorithmic thinking abilities are improved. 8 Computational Intelligence and Neuroscience

Comparison of Collaborative Tinking and Critical
Tinking Tests. Te CT was analyzed from the dimensions of Collaborative Tinking and Critical Tinking. Figure 10 compares the results of the pretest and post-test. In Figure 10(a), there are individual diferences in the thinking of 10 students. Te pretest value of student No. 2 is 2.3, and the post-test value has increased to 2.8, which is the lowest value; student No. 6 has the highest value. Te post-test values of collaborative thinking in all samples are higher than in the pre-test. In the test of critical thinking in Figure 10(b), the pretest value of student No. 1 is 2.7, and the post-test value is increased to 3.2, which is the lowest value; student No. 6 is the highest.
Te post-test values of the critical thinking dimension are higher than the pretest values, and the overall trend also shows an upward trend.

Comparison of Problem-Solving Ability and Self-
Efcacy. Te CT test was performed in terms of problemsolving ability and self-efcacy. Figure 11 presents the results of the pretest and post-test. Figure 11(a) illustrates that the post-test value of 10 students' problem-solving ability is greater than the pretest value. Student No. 1 has the lowest value; the pretest value is 3, and the post-test value is 3.4. Te overall  improvement is about 0.4. Te increase in the proportion is not very obvious, but to a certain extent, it can still improve the students' problem-solving ability. Figure 11(b) refers that the post-test value of most of the students' self-effcacy is higher than the pretest, and the improvement of No. 5 student is not much, and the increase is 0.1. On the whole, students' self-efcacy has a certain improvement compared with the pretest.

Analysis of Cognitive Load.
Te CT test was analyzed from the perspective of cognitive load. Figure 12 compares the results of the pre-test and post-test.
In Figure 12, the post-test values of the cognitive load of the 10 students are all lower than the pretest values, with the most obvious reductions for students No. 4 and No. 5, and the test values decreased by 0.5. Overall, the teaching model based on the idea of scratch modular programming can reduce the cognitive load of each learner to varying degrees.

Comparative Analysis of the Average Data of Students' CT on the Pretest and Post-test.
A paired sample t-test was performed on the self-efcacy and cognitive load of CT, and the results are shown in Figure 13.
In Figure 13, A stands for creative thinking, B for algorithmic thinking, C for collaborative thinking, D for critical thinking, E for problem-solving ability, F for self-efcacy, and G for cognitive load. Te post-test average of each dimension is basically higher than the pretest average of each dimension, and the post-test average of cognitive load is smaller than the pretest. Te teaching model of modular programming idea has an impact on the learning efect of learners. Te P values of algorithmic thinking, collaborative thinking, critical thinking, problem-solving ability, and cognitive load are all less than 0.05, and there are signifcant diferences between the post-test and the pretest results. In the dimension of creative thinking, the P value is 0.218, and in the dimension of self-efcacy, the P value is 0.034. Tere is no obvious diference between the dimension of the two.

Analysis of the DL Evaluation Model.
Te DL efect of college students were analyzed from three dimensions: knowledge and skills, process and methods, and emotional attitudes and values. SPSS25 was used to conduct overall descriptive statistics. Te results are shown in Figure 14.
In Figure 14(a), A1 indicates that only knowledge points have been mastered; A2 indicates that the learned knowledge can be expressed correctly; A3 indicates that the chart and data can be explained; and A4 indicates that the learned knowledge is applied to new situations. In Figure 14(b), B1 means it is fun to learn new knowledge; B2 means that all opinions must be supported by evidence; B3 means to analyze the key points of the problem before solving the problem; and B4 means to be good at systematically planning to solve complex problems. In Figure 14(c), C1 means passing the exam with as little work as possible; C2 means that the course is uninteresting and less study time; C3 means selective learning according to necessity; and C4 means that the study irrelevant to the test is meaningless. Te average value of college students' DL knowledge and skills dimensions is greater than 3.5, approaching 5, indicating that college students can reach a certain standard level in terms of knowledge and skills. Te average value of the process and method dimension is about 3.1 > 3, the minimum value is 2.96, and the maximum value is 3.36, indicating that the overall level of the tested students in terms of process and method is slightly low and needs to be further improved. Finally, in terms of emotional attitudes and values, the average value is 2.77 < 3, the maximum value is 3.22, the minimum value is 2.01, the results are not satisfying. Te mean values of the three dimensions are relatively concentrated, and most of them are between "disagree" and "general," which belong to the description of uncertainty. It can be concluded that the level of DL of college students is relatively low. It is necessary to improve students' DL level from the perspective of knowledge and skills, process, method, emotion, attitude, and values.

Conclusion
To improve the CT ability of students in programming courses, a teaching model based on Scratch modular programming is implemented, and the steps of the modular programming-based teaching model and the visual programming problem-solving model are explored. To evaluate the efectiveness of the teaching model based on the idea of modular programming, a quasiexperimental method is used to conduct CT, CTS, pretest, and post-test questionnaires of the group self-efcacy scale and cognitive load. Furthermore, DL evaluation analysis is carried out on the learning results. Te following conclusions are drawn. Te efectiveness of the teaching model based on scratch modular programming has been verifed on the basis of experiments. Te T-test is performed on the paired samples of students' CT, t � -2.08, P � 0.02. Tere are obvious diferences in the results of pretest and post-test, illustrating that the teaching model based on modular programming can cause changes in students' CT ability. Te post-test values of the dimensions of algorithmic thinking, critical thinking, collaborative thinking, and problem-solving thinking are all higher than the pretest values, and there are individual diferences in each sample. Te P values are all less than 0.05, and the posttest values of cognitive load are all lower than the pretest values. It means that the CT training of the designed teaching model has a certain positive efect on the algorithm dimension, critical thinking, collaborative thinking, and problem-solving ability of students' CT and reduces cognitive load. However, there is no obvious diference in the dimensions of creativity and self-efcacy. On the basis of the DL evaluation, it can be found that college students can reach a certain standard level in terms of knowledge and skills. Te process and methods, as well as emotional attitudes and values, need to be strengthened. Tis study also has certain limitations. Like most educational research experiments, the sample size of this experiment is not very large. Tere are still some limitations in the collection of learner data, and more methods and tools to efectively collect learner thinking data need to be explored. Moreover, individual diferences among learners may have some infuence on the experimental results. Due to the diferences in individual thinking among learners, it is difcult to comprehensively consider and carry out fully targeted training in Scratch courses. Te applicability of the teaching model needs to be further expanded. In the follow-up, there will be targeted training strategies and methods, and diferent levels of practical research will be carried out to improve the universality and efectiveness of the model many times.

Data Availability
Te data used to support the fndings of this study are included within the article.

Conflicts of Interest
Te authors declare that there are no conficts of interest.