Military school student cognitive ability evaluation method based on multi-agent and Bloom education target classification

Abstract

The invention discloses a military school student cognitive ability evaluation method based on multi-agent and Bloom education target classification, and aims to develop an effective method to evaluate the online learning cognitive ability of a student, which is more and more extensive in use of an online learning platform in military colleges and universities. In order to analyze the cognitive level and ability obtained by students in online learning, a multi-agent system is developed in combination with deep learning and a Bloom education target classification theory, the system is composed of a plurality of independent agents, each agent is responsible for evaluating the cognitive ability of the students on different levels of a Bloom classification method, and the cognitive ability of the students on different levels of the Bloom classification method is evaluated. The system processes and analyzes learning data, including homework, test, classroom participation and the like, of trainees by using a deep learning technology, and evaluates the cognitive competence of the trainees through good, poor and average according to six levels of a Bloom classification method. According to the experiment, text replies of 300 students including 12 subjective questions are selected, and cognitive evaluation results of different machine learning algorithms are explored.

Description

Technical Field

[0001] The present invention relates to the technical field of assessment methods, and more particularly, to a method for assessing the cognitive abilities of military cadets based on multi-agent and Bloom's taxonomy of educational objectives. Background Art

[0002] Online courses are now widely used in military academies. To improve online learning outcomes, instructors must promptly assess students' cognitive abilities and provide feedback. Students' cognitive engagement can be used to assess the effectiveness of online learning models and has a significant impact on learning performance and knowledge construction, which is crucial for meaningful learning. Therefore, it is crucial to evaluate students' performance and measure their cognitive proficiency throughout online courses.

[0003] Many techniques have been used to assess learners' performance and cognitive abilities in online learning platforms. Some researchers use automatically created questions to test online learners' real-time cognitive characteristics, collect video text from online courses, and use models to create questions to evaluate answers. Studies by Yang et al. and Lehman et al. used two methods to identify the cognitive complexity of over 2,000 questions. Some methods only allow diagrams as answers to categorize learners' physical cognitive evaluations and then analyze their responses based on risk-based cognitive levels (RBT). Other methods propose using predictive models to investigate learners' intellectual engagement in distance learning. Learners' engagement is tracked through a learning management system (LMS), and written comments are collected during exercises. Finally, the written information is evaluated using a cognitive engagement coding structure, and learners' engagement is analyzed by login activity, access to course materials, and frequency of participation in discussions.

[0004] However, previous research using computer vision to analyze cognitive levels has focused on behavior and emotion. Due to their increasing complexity, the assessment of cognitive levels and abilities has often remained unaddressed. Regarding the relationship between cognition and knowledge, some studies have suggested that students with different cognitive levels and abilities are associated with varying levels of knowledge and mental effort. Furthermore, there are studies using computer text analysis methods to assess the cognitive level of interactive content in online learning forums. Summary of the Invention

[0005] (1) Technical problems solved

[0006] In response to the problems existing in the prior art, the present invention provides a method for evaluating the cognitive abilities of military cadets based on multi-agent and Bloom's taxonomy of educational objectives to solve the technical problems mentioned in the background technology.

[0007] (2) Technical solution

[0008] In order to achieve the above object, the application provides the following technical scheme: a military school student cognitive ability evaluation method based on multi-agent and Bloom education target classification, comprising sequentially connected upload Bloom classification problem sample module, data preprocessing module, Bloom classification level prediction model and student cognitive evaluation oriented agent.

[0009] The application is further provided that the data preprocessing module involves preprocessing the test questions and the text answers of the students using natural language processing (NLP) and converting them into a data set form. Stop words, misspelled words and the like are common in text answers, but these insignificant features will have a negative impact on the performance of machine learning (ML) algorithms. Therefore, it is necessary to use natural language database (NLTK) for text preprocessing first.

[0010] The application is further provided that in the student online learning cognitive evaluation based on the multi-agent system, feature extraction is a crucial step, which determines the degree to which the subsequent machine learning model can understand and utilize the text information of the students. In the system, two feature extraction methods are adopted: term frequency-inverse document frequency (TF-IDF) and Word2Vec, which convert the original text information into numerical vectors that can be understood and processed by machine learning algorithms.

[0011] The application is further provided that the feature extraction package comprises the following steps:

[0012] Firstly, term frequency-inverse document frequency (TF-IDF) is a statistical method widely used in text mining and information retrieval. The importance of a word in the entire corpus is measured by calculating the product of the frequency (TF) of the word in the document and the inverse document frequency (IDF) in the entire corpus. TF represents the frequency of a word in a document, i.e. the ratio of the number of times a word appears in a document to the total number of words in the document. IDF reflects the general importance of a word in all documents. If the number of documents containing the word is smaller, the IDF value is larger, indicating that the word has good class discrimination ability. The calculation of term frequency-inverse document frequency (TF-IDF) is shown in the formula:

[0013] W(d, t) = TF(d, t) * log(N / df(t))

[0014] Where t represents a word and d represents a document. By calculating the TF-IDF value of each word in the document, a TF-IDF vector can be obtained, which can represent the theme and core content of the document. The system uses the TF-IDF vector to extract features from the text answers of the students, which will be used in subsequent machine learning classification, clustering and other tasks.

[0015] On the other hand, Word2Vec is a language model based on neural networks that represents each word as a vector. These vectors can capture the semantic relationship between words. Word2Vec learns contextual information from large amounts of text data, so that semantically similar words are also positioned similarly in the vector space. This system uses the Word2Vec model to convert the words in the students' text answers into vector representations for subsequent semantic analysis and similarity calculation.

[0016] Finally, there are multiple ways to train the Word2Vec model, the most commonly used of which are the Continuous Bag of Words (CBOW) model and the Skip-Gram model. The CBOW model predicts the target word based on the context, while the Skip-Gram model predicts the context based on the target word. This system chooses the CBOW model to train Word2Vec because it is highly efficient and stable when processing large amounts of text data. By combining Word2Vec with the TF-IDF method, more comprehensive features can be extracted from students' text answers, including word frequency, importance, and semantic information. These features will be used in subsequent multi-agent systems to evaluate students' cognitive abilities in online learning.

[0017] The present invention is further configured such that the Bloom's educational objectives classification prediction model

[0018] After preprocessing and feature extraction, the Bloom's Educational Objectives Taxonomy prediction model is used. This step mainly classifies the questions according to the cognitive level so that the relevant intelligent agent can be called in the next module to analyze the text response.

[0019] The level prediction based on Bloom's taxonomy uses a support vector machine, which takes the numerical features of the question as input and outputs a digital code from 1 to 6 to represent the cognitive level;

[0020] In order to classify the data, the support vector machine searches for the optimal hyperplane in high-dimensional space to decompose the decision boundary between the data and divides the data. It maximizes the separation of Bloom's educational objective classification and determines the optimal support vector (the data point closest to the decision boundary) based on the radial basis kernel function.

[0021] The present invention further provides that, after the student cognitive assessment model determines the problem level, the next step is to call the relevant agent to perform text analysis and evaluate the student's performance on the problem. Knowledge-level problems are analyzed by knowledge-level agents, and the same applies to other problems. Six agents were designed and trained on a random forest classifier. Each agent takes as input text response features extracted from term frequency-inverse document frequency (TF-IDF) and Word2Vec, and will produce good, bad, and average results;

[0022] In the model, n_estimators = 600 means that 600 trees are built; max_features = 4 represents the number of feature subsets selected; max_features = 3 means that no more than three features are considered for inclusion in any given tree; bootstrap defaults to True, indicating the use of random forests; random_state is set to 18, which represents the student's performance at different classification levels once the RF model is trained on each specific agent using the text answers to a course question.

[0023] (3) Beneficial effects

[0024] Compared with the existing technology, the present invention provides a method for assessing the cognitive ability of military cadets based on multi-agent and Bloom's taxonomy of educational objectives, which has the following beneficial effects:

[0025] This paper uses Bloom's taxonomy to measure students' cognitive levels and develops a multi-agent system to evaluate the performance of military academy students. This provides a solution to the problem of student performance assessment in an online environment. This research is novel in that it covers subjective questions, whereas previous studies have been limited to objective questions. Different natural language processing (NLP) techniques were used to process data collected from students enrolled in a course. The processed dataset was then used with a support vector machine (SVM) classifier to predict the Bloom's taxonomy of educational objectives to which the questions belonged, with 98% accuracy. After determining the hierarchy of the questions, agents at the corresponding level were invoked to analyze the students' textual responses. A random forest algorithm categorized students' performance into three levels: poor, fair, and good, with 92% accuracy. The results demonstrate that the system can be further expanded to assess students' cognitive skills in other military academy courses or lab assignments. Furthermore, the system can be modified to provide real-time feedback to students. The agents can be trained to compare students' answers with actual answers and identify discrepancies. In the future, the system could be further improved to predict students' psychological states by assessing performance across all subjects, not just a single one. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a framework diagram of the method for assessing the cognitive abilities of military cadets based on multi-agent and Bloom's Taxonomy of Educational Objectives in the present invention;

[0027] Figure 2 Schematic diagram of word frequency-inverse document frequency feature extraction in the present invention;

[0028] Figure 3 This is a schematic diagram of the prediction of the random forest classifier in the present invention;

[0029] Figure 4The confusion matrix schematic diagram of the support vector machine in the application;

[0030] Figure 5 The performance schematic diagram in the application. DETAILED DESCRIPTION

[0031] It should be noted that the embodiments and features in the embodiments in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0032] It should be noted that, unless otherwise specified, all technical and scientific terms used in the present application have the same meaning as that generally understood by those skilled in the art to which the present application belongs.

[0033] In the present application, unless otherwise specified, the orientation such as "upper", "lower" is generally with respect to the direction shown in the drawings, or with respect to the vertical, perpendicular or gravity direction; similarly, for the convenience of understanding and description, "left", "right" is generally with respect to the left and right shown in the drawings; "inner", "outer" refers to the inner and outer with respect to the contour of each component itself, but the above orientation words are not used to limit the present application.

[0034] 1. Method

[0035] To improve the effectiveness and efficiency of evaluating cadets' cognitive ability in an online learning environment, a multi-agent system is designed in this paper. The system is composed of multiple autonomous agents that cooperate with each other to achieve common goals. In the context of cognitive assessment, the multi-agent system can provide personalized assessment and feedback to adapt to different students' learning styles and support the overall assessment process. In addition, each agent can convert text into numerical features, and then measure students' cognitive performance in an online learning environment through text answers.

[0036] 2. System Architecture

[0037] The proposed model is demonstrated with technical components in Figure 1 The model first builds a performance prediction model using Bloom's Taxonomy to classify cognitive levels. All questions based on the six levels of Bloom's Taxonomy are input into the model. The model parses each question by tokenizing each word, and then removes stop words from the question using word stemming. Then, the model uses stem extraction to simplify the most important terms in the query to their most basic form. Finally, the question is converted into a dataset. Important items are used as independent variables, and Bloom's educational goal classification of the question is used as the dependent variable. Compared with other classifiers in the study, support vector machines are widely used in the model and can produce the best prediction accuracy. As shown in Figure 1

[0038] ​The method covers the design of sample questionnaire, data collection, data preprocessing and Bloom taxonomy level prediction model, agent design for student text response analysis, model testing and evaluation. To train the support vector machine (SVM), this paper first collects the problems of a course in an online education platform and previous research work, uses the natural language database (Natural Language ToolKit) in Python to complete the data preprocessing, and assigns a digital code for the Bloom educational goal classification. Then use the term frequency-inverse document frequency (TF-IDF) to convert each question into a feature vector, and use SVM to classify the cognitive level of the question. After the SVM classifier is ready, use python to design a multi-agent system to train each agent for text response analysis. Then use the Beautiful Soup Library to collect the text answer dataset. Then use F-IDF and Word2Vec to extract text features so that NLP models such as random forest classifier can be trained on each agent. After the model is trained, the cognitive level of the collected student response dataset is divided into good, poor and average to evaluate the student's performance at each level.

[0039] 3. System modules

[0040] This section introduces the detailed description of each module.

[0041] (1) Data preprocessing

[0042] This step involves using natural language processing (NLP) to preprocess the text answers of the test questions and students into a dataset form. Stop words, misspelled words, etc. are common in text answers, but these irrelevant features will have a negative impact on the performance of machine learning (ML) algorithms. Therefore, it is necessary to use the natural language database (NLTK) for text preprocessing first.

[0043] Figure 2 Term frequency-inverse document frequency feature extraction

[0044] (2) Feature extraction

[0045] In the online learning cognitive assessment of students based on multi-agent system, feature extraction is a crucial step that determines the extent to which subsequent machine learning models can understand and utilize student text information. In this system, two feature extraction methods are used: term frequency-inverse document frequency (TF-IDF) and Word2Vec, which convert raw text information into numerical vectors that machine learning algorithms can understand and process.

[0046] Firstly, Term Frequency-Inverse Document Frequency (TF-IDF) is a statistical method widely used in text mining and information retrieval. It measures the importance of a word in the entire corpus by calculating the product of its frequency in a document (TF) and its inverse document frequency (IDF). TF represents the frequency of a word in a document, i.e., the number of times a word appears in a document divided by the total number of words in that document. IDF reflects the general importance of a word across all documents, with a lower number of documents containing the word resulting in a higher IDF value, indicating better class discrimination ability. The calculation of TF-IDF is shown in Equation (1):

[0047] W(d, t) = TF(d, t) * log(N / df(t)) (1)

[0048] where t represents a word and d represents a document. By calculating the TF-IDF value of each word in a document, a TF-IDF vector can be obtained, which can represent the theme and core content of the document. The system uses the TF-IDF vector to extract features from the student's text answers, which will be used in subsequent machine learning classification, clustering, and other tasks.

[0049] On the other hand, Word2Vec is a neural network-based language model that represents each word as a vector, which can capture the semantic relationship between words. Word2Vec learns the context information in a large amount of text data, so that semantically similar words are close in vector space. The system uses the Word2Vec model to convert the words in the student's text answers into vector representations for subsequent semantic analysis and similarity calculations.

[0050] There are several training methods for the Word2Vec model, among which the most commonly used are the Continuous Bag-of-Words model (CBOW) and the Skip-Gram model. The CBOW model predicts the target word based on the context, while the Skip-Gram model predicts the context based on the target word. The system selects the CBOW model to train Word2Vec because the CBOW model has higher efficiency and stability when dealing with large amounts of text data. By combining Word2Vec with the TF-IDF method, we can more comprehensively extract features from the student's text answers, including word frequency, importance, and semantic information. These features will be used in subsequent multi-agent systems to evaluate the student's online learning cognitive abilities.

[0051] (3) Bloom's Taxonomy of Educational Objectives Prediction Model

[0052] After preprocessing and feature extraction, the Bloom's Educational Objectives Classification prediction model is used. This step mainly classifies the questions according to the cognitive level so that the relevant intelligent agent can be called in the next module to analyze the text response.

[0053] The hierarchical prediction based on Bloom's taxonomy adopts support vector machine, which takes the numerical features of the question as input and outputs a digital code of 1-6 to represent the cognitive level.

[0054] In order to classify the data, the support vector machine searches for the optimal hyperplane in high-dimensional space to decompose the decision boundary between the data and divides the data. It maximizes the separation of Bloom's educational objective classification and determines the optimal support vector (the data point closest to the decision boundary) based on the radial basis kernel function.

[0055] (4) Trainee cognitive assessment model

[0056] After determining the question hierarchy, the next step is to invoke the relevant agents to perform text analysis and evaluate the learner's performance on that question. Knowledge-level questions are analyzed using knowledge-level agents, and the same applies to other questions. Six agents were designed and trained on a random forest classifier. Each agent takes as input text response features extracted from Term Frequency-Inverse Document Frequency (TF-IDF) and Word2Vec and generates good, bad, and average results.

[0057] In the model, n_estimators=600 means that 600 trees are built; max_features=4 represents the number of feature subsets selected; max_features=3 means that no more than three features are considered for inclusion in any given tree; bootstrap defaults to True, indicating the use of random forest; random_state is set to 18. Once the RF model is trained on each specific agent using the text answers to a course question, it will represent the student's performance at different classification levels, such as Figure 3 shown.

[0058] Figure 3 Random Forest Classifier Prediction

[0059] Based on the answers provided by the students, the proposed method can test the students' performance to evaluate their engagement in the online learning platform.

[0060] experiment

[0061] This section describes the setup of the experiments, which includes the main libraries used with the dataset. Table 2 shows the accuracy comparison of different algorithms, with cosine similarity at 70%, naive Bayes at 92%, and support vector machine at 98%.

[0062] Table 2 Comparison of different classifiers

[0063]

[0064] Experimental setup

[0065] This experiment uses Google Collab and Python. Google Collab is a popular online integrated development environment (IDE) for deep learning and machine learning. The following libraries were also used: Python 3, Pandas, Requests, BeautifulSoup, the Natural Language Processing Toolkit (NLTK), Numpy, Scikit-learn, Gensim, Matplotlib, and Seaborn. Natural language processing-related libraries were used for text analysis. The training and test sets were split 8:2.

[0066] Dataset Collection

[0067] Bloom's Taxonomy has been widely adopted in teaching tasks. This study used online courses to assess students. To meet the learning objectives of a course and ensure effective assessment, evaluators must categorize questions according to different cognitive levels. To this end, an online test created on an online education platform was used. It asked 12 subjective questions, each corresponding to one of the six levels of Bloom's Taxonomy, to measure the students' psychological state. The online test was conducted among students enrolled in the fifth semester of a computer science undergraduate course. To achieve the goals of this article, the test results were treated as a graded task to ensure maximum student engagement. The exam was administered at the end of the online course. The online test consisted of 12 questions and lasted 30 minutes. Text responses from 300 students were collected and used as input for an intelligent system.

[0068] The criteria for determining whether students are good, poor, or average are their scores at each level. Students who score above 60% at the analysis level and above 70% at the application level are considered "good," students who score below 60% at the analysis level are considered "poor," and students who score above 50% at the application level are considered "average."

[0069] Results of different classifiers

[0070] The questions in the online test were designed based on Bloom's Taxonomy, and support vector machines were used to classify the levels of Bloom's Taxonomy. Support vector machines were chosen because recent literature shows that they are the most widely used classifier. Furthermore, compared to other machine learning-based algorithms, support vector machines achieved the best results in terms of accuracy in predicting Bloom's taxonomy of educational objectives, such as Figure 4 shown.

[0071] Figure 4Confusion Matrix for Support Vector Machines

[0072] The confusion matrix provides an additional metric for evaluating SVM performance. The confusion matrix compares the proportion of labels that the model correctly classified with the proportion of incorrect predictions. The diagonal values ​​represent the number of labels that were correctly classified. The confusion matrix for the SVM classifier is shown in Figure 1. Figure 4 The confusion matrix shows values ​​magnified 100 times, as the actual instances are much smaller. The accuracy and kappa statistic of the SVM classifier are compared, as shown in Table 3.

[0073] Table 3 SVM classifier results

[0074]

[0075] After predicting the question level, numerical features of the text response, derived from Term Frequency-Inverse Document Frequency (TF-IDF) and Word2Vec, are fed into the corresponding agent. A machine learning model is then used to generate the results as a performance table. Random forest is one of the best classifiers for evaluating student text performance. Each agent returns the evaluation results for 300 students in the form of good, bad, and average. These results are then combined to determine the overall accuracy of the system, as shown in Table 4.

[0076] Table 4 Random forest experiment results

[0077]

[0078]

[0079] Student performance evaluation

[0080] Figure 5 Performance of students at the knowledge level; performance of students at the application level; performance of students at the analysis level; performance of students at the synthesis level; performance of students at the evaluation level

[0081] like Figure 5 Displays the student's Bloom's taxonomy performance results in a certain course. Figure 5 It shows the knowledge level performance of the students, among which 150 students performed well, 50 were average, and 100 were poor, which shows the students' ability to recall and recognize information. Figure 5 It shows the level of understanding of the students, among which 80 students performed averagely and poorly, while 140 students performed well. Figure 5 The analysis level performance of the students is shown, where 90 students performed well, 123 performed averagely and 87 performed poorly. Figure 5 Shows students' performance at the comprehensive level. Students' performance at the more difficult comprehensive levels is not satisfactory. This provides students with opportunities to improve their performance and enhance their learning in order to compete with the comprehensive level and other cognitive levels of the taxonomy. Figure 5 It shows the performance of the students in the evaluation level. The evaluation level requires students to make judgments based on standards and criteria. Through inspection and evaluation, due to the complexity of cognitive level, the students' main performance in the evaluation level is unsatisfactory.

[0082] Table 4 shows the precision, recall, F1, and accuracy of the Bloom’s taxonomy at different levels for the students, with an overall accuracy of 91.83%.

[0083] When learners' strengths and weaknesses are visualized using a bar chart at each level of Bloom's taxonomy, their performance becomes more readily apparent. The results indicate that most learners perform well on knowledge and application questions, but only a minority perform well on synthesis and evaluation questions. By understanding these performance patterns, educators can adjust their teaching strategies appropriately. For example, teachers should prioritize helping learners acquire higher-order thinking skills to address gaps and enhance overall cognitive learning. This targeted teaching approach will help learners develop the necessary skills at all levels of cognitive complexity, leading to greater academic success.

[0084] While 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 these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the cognitive abilities of military cadets based on multi-agent and Bloom's taxonomy of educational objectives is characterized by: It includes a module for uploading Bloom's taxonomy problem samples, a data preprocessing module, a Bloom's taxonomy hierarchical prediction model and an intelligent agent for student cognitive assessment, which are connected in sequence.

2. The method for assessing the cognitive abilities of military cadets based on multi-agent and Bloom's taxonomy of educational objectives according to claim 1 is characterized by: The data preprocessing module involves using natural language processing (NLP) to preprocess the test questions and students' text answers and convert them into a dataset. Stop words, misspellings, etc. are common in text answers, but these insignificant features will have a negative impact on the performance of machine learning algorithms. Therefore, it is necessary to use the natural language database NLTK for text preprocessing.

3. The method for assessing the cognitive ability of military cadets based on multi-agent and Bloom's taxonomy of educational objectives according to claim 2 is characterized by: In the cognitive assessment of online learning by multi-agent systems, feature extraction is a crucial step that determines the extent to which subsequent machine learning models can understand and utilize students' text information. In this system, two feature extraction methods are used: Term Frequency-Inverse Document Frequency (TF-IDF) and Word2Vec. These two methods convert raw text information into numerical vectors that can be understood and processed by machine learning algorithms.

4. The method for assessing the cognitive abilities of military cadets based on multi-agent and Bloom's taxonomy of educational objectives according to claim 3 is characterized by: The feature extraction package includes the following steps: First, term frequency-inverse document frequency (TF-IDF) is a statistical method widely used in text mining and information retrieval. It measures the importance of a word in the entire corpus by calculating the product of the frequency (TF) of a word in a document and the inverse document frequency (IDF) of a word in the entire corpus. TF represents the frequency of a word in a document, that is, the ratio of the number of times a word appears in a document to the total number of words in the document. IDF reflects the general importance of a word in all documents. The fewer documents containing the word, the larger the IDF value, indicating that the word has good category discrimination ability. The calculation formula of term frequency-inverse document frequency (TF-IDF) is shown as follows: W(d,t)=TF(d,t)*log(N / df(t)) Here, t represents a word and d represents a document. By calculating the TF-IDF value of each word in a document, we can obtain a TF-IDF vector, which can represent the theme and core content of the document. This system uses the TF-IDF vector to extract features from the students' text answers. These features will be used in subsequent machine learning tasks such as classification and clustering. On the other hand, Word2Vec is a language model based on neural networks that represents each word as a vector. These vectors can capture the semantic relationship between words. Word2Vec learns contextual information from large amounts of text data, so that semantically similar words are also located close together in the vector space. This system uses the Word2Vec model to convert the words in the students' text answers into vector representations for subsequent semantic analysis and similarity calculation. Finally, there are many ways to train the Word2Vec model, the most commonly used of which are the continuous bag-of-words model (CBOW) and the skip-gram model (Skip-Gram). The CBOW model predicts the target word based on the context, while the Skip-Gram model predicts the context based on the target word. This system chooses the CBOW model to train Word2Vec because the CBOW model has high efficiency and stability when processing large amounts of text data. By combining Word2Vec with the TF-IDF method, the features in the students' text answers can be more comprehensively extracted, including the frequency, importance, and semantic information of the vocabulary. These features will be used in subsequent multi-agent systems to evaluate the students' online learning cognitive ability.

5. The method for assessing the cognitive abilities of military cadets based on multi-agent and Bloom's taxonomy of educational objectives according to claim 4 is characterized by: The Bloom's Taxonomy of Educational Objectives prediction model After preprocessing and feature extraction, the Bloom's Educational Objectives Taxonomy prediction model is used. This step mainly classifies the questions according to the cognitive level so that the relevant intelligent agent can be called in the next module to analyze the text response. The level prediction based on Bloom's taxonomy uses a support vector machine, which takes the numerical features of the question as input and outputs a digital code from 1 to 6 to represent the cognitive level; In order to classify the data, the support vector machine searches for the optimal hyperplane in high-dimensional space to decompose the decision boundary between the data and divides the data, by maximizing the separation of Bloom's educational objective classification and determining the optimal support vector based on the radial basis kernel function.

6. The method for assessing the cognitive abilities of military cadets based on multi-agent and Bloom's taxonomy of educational objectives according to any one of claims 1 to 5, wherein: After the student cognitive assessment model determines the problem level, the next step is to call the relevant intelligent agent to perform text analysis and evaluate the student's performance on the problem. Knowledge-level problems were analyzed using knowledge-level agents. Similarly, six agents were designed and trained on a random forest classifier. Each agent took text response features extracted from TF-IDF and Word2Vec as input and produced good, bad, and average results. In the model, n_estimators = 600 means that 600 trees are built; max_features = 4 represents the number of feature subsets selected; max_features = 3 means that no more than three features are considered for inclusion in any given tree; bootstrap defaults to True, indicating the use of random forests; random_state is set to 18, which represents the student's performance at different classification levels once the RF model is trained on each specific agent using the text answers to a course question.

Images