College student artificial intelligence course online evaluation method and related equipment

Abstract

The invention relates to a college student artificial intelligence course online evaluation method and related equipment, and the method comprises the following steps: carrying out the construction of a knowledge point dependency chain of a course outline knowledge point, and obtaining a knowledge point dependency chain; obtaining an online behavior log of a target user, and inferring a learning track of the target user in the knowledge point dependency chain based on the online behavior log through a preset hidden Markov model; based on the learning track and the online behavior log, predicting a mastering probability of a target user at each course outline knowledge point; performing weight adjustment on each course outline knowledge point based on the mastering probability to obtain an adjusted knowledge point weight; and evaluating the online condition of the target user based on the mastering probability corresponding to the knowledge points of the course outline and the adjustment of the weight of the knowledge points, thereby solving the technical problem that a traditional evaluation method mainly depends on an end-of-period test or a regular test and often cannot reflect the learning progress and the knowledge mastering condition of students in real time.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an online evaluation method for an artificial intelligence course for college students and related equipment. Background Art

[0002] The rapid development of online education has made the evaluation method of artificial intelligence courses an urgent issue to be solved. Traditional evaluation methods mainly rely on final exams or regular tests, which often cannot reflect students' learning progress and knowledge mastery in real time, especially in a field like artificial intelligence that covers a wide range of knowledge points and complex logical relationships. With the popularization of online learning platforms, students' learning behavior data has become more abundant and traceable, but how to effectively use this data to accurately evaluate students' learning effects remains a challenge.

[0003] At present, many online course evaluation methods ignore the dependencies between knowledge points, resulting in inaccurate evaluation results. For example, a student may perform well on a certain knowledge point, but because he has not mastered its pre-dependent knowledge points, the actual learning effect is not as shown in the evaluation. At the same time, traditional methods often use unified evaluation standards, which cannot reflect each student's learning path and mastery level in a personalized way, and cannot provide targeted improvement suggestions. In addition, when processing large-scale online behavior data, existing evaluation systems often face problems of high computational complexity and poor real-time performance, making it difficult to meet the needs of instant feedback in online learning.

[0004] Therefore, it is particularly important to study a method that can dynamically, individually and accurately evaluate students' learning effects. This method needs to be able to build a knowledge point dependency chain, identify students' learning trajectories, and use advanced statistical methods such as hidden Markov models to predict students' mastery probability. This evaluation method can not only provide more accurate learning evaluation, but also help teachers and learning platforms optimize teaching content and strategies, thereby improving learning efficiency and solving many pain points in current online education evaluation. Summary of the invention

[0005] The main purpose of the present invention is to provide an online evaluation method and related equipment for college students' artificial intelligence courses, which solves the technical problem that traditional evaluation methods mainly rely on final exams or regular tests and often cannot reflect students' learning progress and knowledge mastery in real time.

[0006] To achieve the above object, the present invention provides an online evaluation method for an artificial intelligence course for college students, comprising the following steps: Acquire the syllabus knowledge points of the artificial intelligence course, and construct a knowledge point dependency chain for the syllabus knowledge points to obtain a knowledge point dependency chain; wherein the knowledge point dependency chain includes the logical relationship between the pre-dependent knowledge points and the post-associated knowledge points of each knowledge point; Obtaining the online behavior log of the target user, and inferring the learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log through a preset hidden Markov model; Based on the learning trajectory and the online behavior log, predict the probability of the target user mastering each knowledge point in the course outline; Adjust the weight of each knowledge point in the course outline based on the mastery probability to obtain an adjusted knowledge point weight; The online status of the target user is evaluated based on the mastery probability corresponding to the knowledge points in the course outline and the weight of the knowledge points is adjusted.

[0007] Furthermore, the knowledge point dependency chain is constructed for the course outline knowledge points to obtain the knowledge point dependency chain, including: Performing semantic analysis on the course outline knowledge points through a preset natural language processing technology to obtain a set of analyzed knowledge points; Performing relationship analysis on the knowledge points in the parsed knowledge point set to establish a dependency graph between the knowledge points; Vectorizing the dependency graph to map each knowledge point in the course outline into a high-dimensional space to obtain a vectorized representation of the knowledge point; Performing modular analysis on the vectorized representation of the knowledge points, identifying closely related subsets between knowledge points in each syllabus, and obtaining a knowledge point module dependency chain; By using a path importance evaluation method, the knowledge point module dependency chain is screened to obtain a screening key path; The graph convolutional network algorithm is used to model the transfer effect between the knowledge points in the critical screening path to obtain the knowledge point dependency chain.

[0008] Furthermore, the inferring the learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log by using a preset hidden Markov model includes: The behavior data in the online behavior log is processed into blocks at preset fixed time intervals to obtain block-based behavior data blocks; wherein the behavior data includes login time data, login times data, video viewing time data, discussion area posting data, and homework submission data; Using wavelet transform technology, extracting frequency features from each of the divided behavior data blocks to obtain an extracted frequency vector; Using a hidden Markov model, based on the extracted frequency vector, a state sequence inference is performed on the target user to obtain a learning state sequence; Based on the learning state sequence and the knowledge point dependency chain, inferring the learning position of the target user in the knowledge point dependency chain; Perform future trend analysis on the learning position to predict the target user's accessed knowledge points in the knowledge point dependency chain, and obtain a learning trajectory based on the accessed knowledge points, online behavior logs and the learning position.

[0009] Furthermore, the predicting the target user's mastery probability of each knowledge point in the course outline based on the learning trajectory and the online behavior log includes: Extracting time series features from the learning trajectory to obtain time series features of the learning trajectory; Encoding the temporal features of the learning trajectory through a preset attention transformer network to obtain a semantic vector of the learning trajectory; Using the Bayesian probabilistic graphical model, the learning trajectory semantic vector and the online behavior log are used to perform probability reasoning on the target user at each knowledge point in the course outline to obtain a learning behavior probability distribution matrix; wherein the learning behavior probability distribution matrix includes the conditional probability and transition probability of knowledge point learning; Perform memory enhancement reasoning on each knowledge point in the course outline based on the learning behavior probability distribution matrix to obtain a dynamic prediction sequence for mastering the knowledge points; Based on the knowledge point mastery dynamic prediction sequence, each of the course outline knowledge points is estimated to have a mastery probability interval, and the mastery probability upper and lower bounds are obtained; wherein the mastery probability upper and lower bounds include a mastery knowledge lower bound and a mastery knowledge upper bound; Based on the upper and lower bounds of the mastery probability, the mastery probability of the target user at each knowledge point in the course outline is predicted.

[0010] Furthermore, the memory enhancement reasoning is performed on each knowledge point in the course outline based on the learning behavior probability distribution matrix to obtain a dynamic prediction sequence of knowledge point mastery, including: Through a memory-enhanced neural network, dynamically reconstructing each of the course outline knowledge points based on the learning behavior probability distribution matrix to obtain memory-reconstructed knowledge points; Performing probabilistic semantic deconstruction on the memory reconstruction knowledge points to obtain a latent semantic space of the knowledge points; Performing path weighted reasoning on the learning behavior probability based on the knowledge point latent semantic space to obtain a knowledge point relevance reasoning graph; Performing probability path inference on the knowledge point association reasoning graph to obtain a knowledge point mastery probability evolution sequence; Based on the probability evolution sequence of mastering the knowledge points, time series prediction is performed on the knowledge points in the course outline to obtain a dynamic prediction sequence of mastering the knowledge points.

[0011] Furthermore, the weight of each knowledge point in the course outline is adjusted based on the mastery probability to obtain the adjusted knowledge point weight, including: Based on the mastery probability, information entropy covariance analysis is performed on each knowledge point in the course outline to obtain a knowledge point complexity tensor; Based on the knowledge point complexity tensor, the course outline knowledge points are subjected to probabilistic semantic association deconstruction to obtain a knowledge point topological association graph; Based on a preset graph attention network, conditional probability reasoning is performed on the knowledge point topological association graph to obtain a probability distribution of knowledge point association; Based on the probability distribution of the knowledge point relevance, a priori probability distribution modeling is performed on the knowledge points in the course outline to obtain a priori distribution of the knowledge point weights; Adaptively nonlinearly mapping the prior distribution of the knowledge point weights to obtain a knowledge point dynamic weight evolution sequence; wherein the knowledge point dynamic weight evolution sequence includes an attenuation coefficient of the knowledge point weight and a learning evolution trajectory; The weight of each knowledge point in the course outline is adjusted based on the dynamic weight evolution sequence of the knowledge point to obtain the adjusted knowledge point weight.

[0012] Furthermore, the online status of the target user is evaluated based on the adjusted knowledge point weights and the mastery probability of the course outline knowledge points corresponding to the adjusted knowledge point weights, including: Through the preset Gaussian mixture model, the joint distribution of the adjusted knowledge point weights and mastery probabilities is estimated to obtain the joint distribution structure data; Randomly sampling the joint distribution structure data by using a Monte Carlo simulation method to obtain virtual sample data; Performing pattern recognition on the virtual sample data to mine hidden potential patterns in the virtual sample data to obtain a potential pattern set; Inputting the potential pattern set into a preset support vector machine model for classification to obtain classification results; wherein the classification results include excellent learning state, medium learning state and poor learning state; The online status of the target user is determined and evaluated based on the classification result.

[0013] The present invention also provides an online evaluation device for college students' artificial intelligence courses, comprising: A construction module is used to obtain the syllabus knowledge points of the artificial intelligence course, and construct a knowledge point dependency chain for the syllabus knowledge points to obtain a knowledge point dependency chain; wherein the knowledge point dependency chain includes the logical relationship between the pre-dependent knowledge points and the post-associated knowledge points of each knowledge point; An inference module, used to obtain the online behavior log of the target user, and infer the learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log through a preset hidden Markov model; A prediction module, used to predict the probability of the target user mastering each knowledge point in the course outline based on the learning trajectory and the online behavior log; An adjustment module, used for adjusting the weight of each knowledge point in the course outline based on the mastery probability to obtain an adjusted knowledge point weight; An evaluation module is used to evaluate the online status of the target user based on the mastery probability corresponding to the knowledge points in the course outline and the weight of the adjusted knowledge points.

[0014] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0015] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0016] The method for online evaluation of an artificial intelligence course for college students provided by the present invention comprises the following steps: obtaining syllabus knowledge points of an artificial intelligence course, and constructing a knowledge point dependency chain for the syllabus knowledge points to obtain a knowledge point dependency chain; obtaining an online behavior log of a target user, and inferring a learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log through a preset hidden Markov model; predicting the mastering probability of the target user in each syllabus knowledge point based on the learning trajectory and the online behavior log; adjusting the weight of each syllabus knowledge point based on the mastering probability to obtain an adjusted knowledge point weight; and evaluating the online status of the target user based on the mastering probability and the adjusted knowledge point weight corresponding to the syllabus knowledge point. By means of the above technical means, the technical problem that the traditional evaluation method mainly relies on final exams or regular tests and often cannot reflect the learning progress and knowledge mastering status of students in real time is solved, and the real-time performance of the online behavior log is utilized. The evaluation system can provide instant feedback, help students adjust their learning strategies in time, and improve the beneficial effect of learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the steps of an online evaluation method for an artificial intelligence course for college students in one embodiment of the present invention; Figure 2 It is a structural block diagram of an online evaluation device for artificial intelligence courses for college students in one embodiment of the present invention; Figure 3It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0018] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] like Figure 1 As shown, Figure 1 This is a schematic diagram of the steps of an online evaluation method for an artificial intelligence course for college students in one embodiment of the present invention; In one embodiment of the present invention, a method for online evaluation of an artificial intelligence course for college students is provided, comprising the following steps: Step S1, obtaining the syllabus knowledge points of the artificial intelligence course, and constructing a knowledge point dependency chain for the syllabus knowledge points to obtain a knowledge point dependency chain; wherein the knowledge point dependency chain includes the logical relationship between the preceding dependent knowledge points and the following associated knowledge points of each knowledge point.

[0021] Specifically, when conducting "an online evaluation method for college students' artificial intelligence courses", it is first necessary to obtain the syllabus knowledge points of the artificial intelligence course, which means extracting all the knowledge points involved from the syllabus. These knowledge points may be algorithms, data structures, machine learning models, programming languages, and other aspects. Next, the knowledge point dependency chain is constructed for the syllabus knowledge points. The construction process is to logically connect these knowledge points according to the dependency relationship between them to form a network diagram of knowledge points. This network diagram not only shows the pre-dependent knowledge points between knowledge points, that is, the knowledge points that need to be mastered before learning a certain knowledge point, but also includes the post-associated knowledge points, that is, the knowledge points or application scenarios that will be involved after learning a certain knowledge point. For example, in the artificial intelligence course, if you want to learn the knowledge point of "deep learning", then its pre-dependent knowledge points may include "linear algebra", "probability statistics", "basics of neural networks", etc., because these are the theoretical foundations necessary to understand deep learning algorithms. At the same time, the learning of deep learning may also be related to post-knowledge points, such as "convolutional neural network (CNN)", "recurrent neural network (RNN)", etc., which are knowledge points that are applied or expanded on the basis of deep learning. By constructing this knowledge point dependency chain, we can clearly see the learning path of each knowledge point, understand the obstacles that students may encounter in the learning process and the knowledge points that need to be strengthened. In practical applications, suppose a student is studying an artificial intelligence course. Through the learning behavior log of the online platform, we can see that he has mastered the basics of linear algebra and probability statistics, and is now learning the basic concepts of neural networks. Using the knowledge point dependency chain, we can predict the problems he may encounter next, such as not having a deep understanding of the "back propagation algorithm", which may be because he has not fully mastered the "chain rule", a pre-dependent knowledge point. Based on this dependency relationship, the system can prompt students to review or strengthen relevant pre-requisite knowledge points, or provide targeted learning resources to help students understand and master the relevant content of deep learning more smoothly. In this way, students' learning efficiency and effectiveness will be significantly improved.

[0022] Step S2, obtaining the online behavior log of the target user, and inferring the learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log through a preset hidden Markov model.

[0023] Specifically, when conducting "an online evaluation method for college students' artificial intelligence courses", obtaining the online behavior log of the target user is a key step. This means that it is necessary to collect all the interactive data of students on the online platform during the process of learning artificial intelligence courses, including access time, click behavior, length of time watching videos, frequency of participating in discussions, records of homework submission, and test scores. These data constitute the online behavior log of the target user and are an important basis for understanding the student's learning situation. Next, the system analyzes these online behavior logs through a preset hidden Markov model. The hidden Markov model is a statistical model used to infer hidden state sequences from observed data. In this case, the observed data is the student's online behavior log, and the hidden state is the student's learning trajectory in the knowledge point dependency chain. Through this model, the system can infer the student's mastery of knowledge points and learning paths in the learning process based on his online behavior. For example, if a student frequently visits related video content and participates in discussions when learning "Basics of Neural Networks", this may mean that he has a high interest in this knowledge point or has certain questions. By analyzing these behavior logs, the system can infer that he may need more time to understand this knowledge point or need additional resource support. Based on the student's online behavior log and the logical relationship of the knowledge point dependency chain, the hidden Markov model predicts the student's learning status at each knowledge point. For example, if a student jumps directly to the study of "deep learning" after learning "machine learning basics" without fully understanding the prerequisite knowledge point "linear algebra", the model may predict that he will encounter obstacles in the process of deep learning. Based on this prediction, the system will adjust the learning path, provide relevant prerequisite knowledge point resources, or prompt students to review or strengthen certain knowledge points. For example, suppose a student is studying an artificial intelligence course, and his online behavior log shows that he often watches video content about "convolutional neural network (CNN)", but rarely participates in related discussions or submits related homework. Through the analysis of the hidden Markov model, the system may infer that he may have a good theoretical understanding of CNN, but his practical application ability may be insufficient. Based on this learning trajectory, the system can recommend more practical cases or provide practical learning resources to help him better master the application of CNN. In this way, the system can understand the student's learning progress and mastery in real time, provide each student with personalized learning paths and support, and improve learning efficiency and effectiveness.

[0024] Step S3: predicting the target user's mastery probability of each knowledge point in the course outline based on the learning trajectory and the online behavior log.

[0025] Specifically, when conducting "an online evaluation method for college students' artificial intelligence courses", it is a key step to predict the target user's mastery probability of each course outline knowledge point based on the learning trajectory and the online behavior log. This process involves comprehensive analysis of the student's online behavior log and learning trajectory to determine his mastery of each knowledge point. First, the student's learning trajectory is analyzed through the hidden Markov model, and the model can identify the student's learning path in the knowledge point dependency chain. The hidden Markov model regards the student's online behavior as an "observation state" and the student's learning state (such as understanding, mastery, need for review, etc.) as a "hidden state". By observing the student's behavior pattern, the model can infer the student's learning state on each knowledge point. For example, if a student frequently accesses learning resources related to "support vector machine (SVM)", but the accuracy of the submitted homework is not high, the model may predict that he has a low degree of mastery in this regard. Next, combined with the online behavior log, the system will analyze the student's interaction, such as access frequency, degree of participation in discussion, test scores, etc. These data provide specific manifestations of students' learning behavior and can be used as a basis for predicting the probability of mastery. For example, if a student participates in many discussions while learning "decision trees" and performs well in related quizzes, the system can judge that he has a high probability of mastering "decision trees". Based on these analyses, the system calculates a mastery probability for each knowledge point, which reflects the student's learning outcomes and understanding of the knowledge point. The high or low mastery probability not only reflects the student's current mastery, but also predicts which knowledge points he may need to invest more energy in. For example, suppose a student is studying an artificial intelligence course, and his online behavior log shows that he has completed the study of "Machine Learning Basics" and scored high in related quizzes. At the same time, he often visits video content about "neural networks" and actively participates in discussions in the discussion area. The hidden Markov model combined with these behavioral data can predict that his mastery probability of "Machine Learning Basics" is high, while the mastery probability of "neural networks" may need to be further improved because he has more interactions in this area, but may need more practice or in-depth learning to consolidate the knowledge. In this way, the system can provide students with personalized learning suggestions, point out which knowledge points they need to strengthen their study or review, and adjust the learning path and evaluation criteria based on the probability of mastery, making the learning process more efficient and targeted.

[0026] Step S4, adjusting the weight of each knowledge point in the course outline based on the mastery probability to obtain an adjusted knowledge point weight.

[0027] Specifically, when conducting "an online evaluation method for college students' artificial intelligence courses", the weight of each knowledge point in the course outline is adjusted based on the mastery probability to obtain the adjusted knowledge point weight, in order to make the evaluation more in line with the students' actual learning situation. The mastery probability reflects the students' learning outcomes and understanding of each knowledge point, and the weight adjustment allows the system to assign different importance to different knowledge points according to the students' actual mastery. First, the system analyzes the students' mastery probability on each knowledge point, which is calculated by the hidden Markov model based on the students' learning trajectory and online behavior log. The level of mastery probability reflects the students' mastery of the knowledge point during the learning process. If the student has a high probability of mastering a certain knowledge point, the weight of the knowledge point may be appropriately reduced because the student is relatively familiar with the content. On the contrary, if the student shows a low mastery probability on a certain knowledge point, the weight of the knowledge point will be increased, which means that the student needs to invest more learning energy and time in this knowledge point. The process of weight adjustment can be regarded as a dynamic balance. In the evaluation, if the student has a high probability of mastering most of the knowledge points, the system will automatically reduce the weight of these knowledge points, so that the evaluation focuses more on the knowledge points that the student needs to strengthen. For example, if a student performs well in learning the relevant knowledge of "artificial neural network", the system may reduce the weight of the knowledge point of "artificial neural network" to encourage him to focus on learning other knowledge points that need to be strengthened, such as "support vector machine" or "Bayesian network". The weight adjustment also takes into account the dependencies between knowledge points. If a knowledge point is a pre-dependent knowledge point of many other knowledge points, then even if the student has a high probability of mastering this knowledge point, the system may not significantly reduce its weight because it is important in the entire learning system. For example, "linear algebra" is the basis of many artificial intelligence algorithms. If the student has mastered "linear algebra", the system will adjust its weight, but will not completely ignore it because it is essential to understand other knowledge points. For example, suppose a student is studying an artificial intelligence course, and his mastery probability shows that he has a high degree of mastery in "machine learning foundations", but his understanding of "deep learning" is relatively weak. Based on these mastery probabilities, the system will adjust the weight of "machine learning foundations" to slightly reduce it, and increase the weight of "deep learning", prompting students to invest more learning resources and time in deep learning. Doing so can not only help students better understand and master knowledge, but also provide teachers with targeted teaching suggestions, optimize course structure, and improve teaching effectiveness. Through this dynamic weight adjustment, the evaluation system can more accurately reflect students' learning status and provide strong support for personalized learning.

[0028] Step S5, evaluating the online status of the target user based on the mastery probability corresponding to the knowledge points in the course outline and adjusting the weights of the knowledge points.

[0029] Specifically, when conducting "an online evaluation method for college students' artificial intelligence courses", the core of the final evaluation step is to evaluate the online situation of the target user based on the mastery probability corresponding to the knowledge points in the course outline and adjust the weight of the knowledge points. Through this step, the system can comprehensively consider the students' mastery of knowledge points and learning focus to provide a comprehensive online learning evaluation. First, the system uses a hidden Markov model to analyze the students' online behavior logs to infer the mastery probability of each knowledge point. This probability reflects the degree of students' understanding and mastery of each knowledge point. The high or low mastery probability not only reflects the student's current learning status, but also predicts which knowledge points he may need more learning time or resources on. Then, the system will adjust the weights of the knowledge points in the course outline according to these mastery probabilities. Adjusting the weight of knowledge points means dynamically adjusting the importance of different knowledge points according to the students' actual learning situation. If the student has a high probability of mastering a certain knowledge point, the weight of the knowledge point may be reduced because the student is relatively familiar with the content. On the contrary, if the student shows a low mastery probability on a certain knowledge point, the weight of the knowledge point will be increased, prompting the student to invest more learning energy in this knowledge point. During the evaluation process, the system comprehensively considers the mastery probability and adjusted weight of each knowledge point to calculate the online learning situation of the target user. The mastery probability indicates the student's mastery of the knowledge point, while the adjusted weight reflects the relative importance of each knowledge point in the student's learning path. For example, if a student has a high mastery probability on "Basics of Neural Networks" but a low mastery probability on "Convolutional Neural Networks (CNN)", and after the weight adjustment, "Convolutional Neural Networks" is given a higher weight, then the system will judge that the student's learning in "Convolutional Neural Networks" needs to be strengthened, and thus reflect this in the final evaluation. This evaluation method not only focuses on the student's mastery of the knowledge points, but also considers the logical relationship between the knowledge points and the learning path. By dynamically adjusting the weights, the system can provide personalized learning suggestions and evaluation results for each student. For example, if a student shows a weak understanding of "Support Vector Machine (SVM)" during the learning process, and the system predicts that he will learn "Bayesian Network" next, then the evaluation system will prompt the student to strengthen his understanding of "Support Vector Machine" before learning "Bayesian Network" because there is a logical relationship between the two knowledge points. Through this comprehensive evaluation, the system can accurately reflect students' learning progress, provide targeted learning suggestions, help students better plan their learning paths, and provide teachers with a basis for optimizing teaching content and strategies. Ultimately, this evaluation method can improve the efficiency and effectiveness of online learning and ensure that students' learning outcomes in artificial intelligence courses are scientifically and objectively evaluated.

[0030] In a specific embodiment, constructing a knowledge point dependency chain for the course outline knowledge points to obtain a knowledge point dependency chain includes: Performing semantic analysis on the course outline knowledge points through a preset natural language processing technology to obtain a set of analyzed knowledge points; Performing relationship analysis on the knowledge points in the parsed knowledge point set to establish a dependency graph between the knowledge points; Vectorizing the dependency graph to map each knowledge point in the course outline into a high-dimensional space to obtain a vectorized representation of the knowledge point; Performing modular analysis on the vectorized representation of the knowledge points, identifying closely related subsets between knowledge points in each syllabus, and obtaining a knowledge point module dependency chain; By using a path importance evaluation method, the knowledge point module dependency chain is screened to obtain a screening key path; The graph convolutional network algorithm is used to model the transfer effect between the knowledge points in the critical screening path to obtain the knowledge point dependency chain.

[0031] Specifically, first, the course outline knowledge points are semantically parsed by using a preset natural language processing technology to obtain a parsed knowledge point set. Natural language processing (NLP) technology plays a key role here. It can analyze the text description in the course outline and extract the core content of each knowledge point. For example, in the artificial intelligence course, NLP can identify the knowledge point of "neural network", including its related concepts, principles, application scenarios, etc. These knowledge points are parsed and formed into a set for subsequent analysis. Next, the knowledge points in the parsed knowledge point set are analyzed in terms of relationships to establish a dependency graph between the knowledge points. Relationship analysis involves identifying pre-dependencies and post-associations between knowledge points. For example, in the artificial intelligence course, "deep learning" depends on "basics of neural networks" because students need to master the basic concepts and principles of neural networks before learning deep learning. By analyzing these relationships, the system can construct a graph in which nodes represent knowledge points and edges represent dependencies between knowledge points. In order to further process these relationships, the system vectorizes the dependency graph. Vectorization is the process of mapping each course outline knowledge point into a high-dimensional space. Through this mapping, each knowledge point is represented as a high-dimensional vector that captures the similarities and relationships between knowledge points. For example, if two knowledge points often appear together or are closely related in the course, their vectors will be close in the high-dimensional space. This provides a mathematical basis for subsequent analysis. Modular analysis of the vectorized representation of knowledge points is a step to identify closely related subsets between knowledge points in each course outline. Modular analysis can discover clustering phenomena between knowledge points. For example, "decision tree", "random forest" and "gradient boosting decision tree" may form a module because they belong to the same class of algorithms in machine learning. Through this analysis, the system is able to identify knowledge point module dependency chains that reflect closer connections between knowledge points. Next, the knowledge point module dependency chain is path-screened by the path importance evaluation method to obtain the screening key path. Path importance evaluation is a method to evaluate the strength and importance of the relationship between knowledge points. For example, learning "convolutional neural network (CNN)" may be an important prerequisite for understanding "object detection", so the importance of this path will be higher. The purpose of screening the key path is to identify the sequence of knowledge points that has the greatest impact on students' learning outcomes during the learning process. Finally, the graph convolutional network (GCN) algorithm is used to model the transmission effect between knowledge points in the screening key path and obtain the knowledge point dependency chain. Graph convolutional network is a deep learning model based on graph structure, which can learn the relationship and information transmission between nodes (here is knowledge points). GCN can simulate the propagation effect of knowledge points on the learning path and predict the problems that students may encounter in the learning process.For example, if a student does not have a good grasp of "Linear Algebra", the GCN model can predict that he will have difficulty learning "Support Vector Machine (SVM)" because linear algebra is the basis of SVM. For example, suppose we have an artificial intelligence course, and the course outline includes knowledge points such as "Machine Learning Basics", "Neural Networks", "Deep Learning", and "Convolutional Neural Networks (CNN)". Through the above steps, the system first parses these knowledge points through NLP, and then builds a dependency graph through relationship analysis, and finds that "Deep Learning" depends on "Neural Networks", and "Convolutional Neural Networks (CNN)" depends on "Deep Learning". Then, these knowledge points are vectorized and represented. In high-dimensional space, "Neural Networks" and "Deep Learning" are close, while "Decision Trees" and "Neural Networks" are far away. Modular analysis may identify that the "Neural Network" series of knowledge points form one module, and the "Decision Tree" series of knowledge points form another module. Through path importance evaluation, the system may find that the path of learning "Convolutional Neural Networks (CNN)" after learning "Deep Learning" is particularly important. Finally, the GCN model simulates the propagation effect of these knowledge points on the learning path and predicts possible obstacles and knowledge points that need to be strengthened in the student learning process. Through these steps, the system can build a knowledge point dependency chain that reflects the logical relationship between knowledge points in the course outline, providing a scientific and systematic basis for online evaluation. By modeling and analyzing the knowledge point dependency chain, the system can not only better understand the student's learning progress, but also provide targeted learning suggestions to improve teaching efficiency and student learning outcomes.

[0032] In a specific embodiment, the inferring the learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log by using a preset hidden Markov model includes: The behavior data in the online behavior log is processed into blocks at preset fixed time intervals to obtain block-based behavior data blocks; wherein the behavior data includes login time data, login times data, video viewing time data, discussion area posting data, and homework submission data; Using wavelet transform technology, extracting frequency features from each of the divided behavior data blocks to obtain an extracted frequency vector; Using a hidden Markov model, based on the extracted frequency vector, a state sequence inference is performed on the target user to obtain a learning state sequence; Based on the learning state sequence and the knowledge point dependency chain, inferring the learning position of the target user in the knowledge point dependency chain; Perform future trend analysis on the learning position to predict the target user's accessed knowledge points in the knowledge point dependency chain, and obtain a learning trajectory based on the accessed knowledge points, online behavior logs and the learning position.

[0033] Specifically, first, the behavior data in the online behavior log is processed in blocks according to a preset fixed time interval. This means that the system divides the student's online learning behavior data (including login time data, login times data, video viewing time data, discussion area posting data, and homework submission data) according to a certain time interval. For example, if one hour is set as a time interval, the behavior data per hour will be regarded as a data block. This can help the system identify the learning behavior patterns of students in different time periods. Next, the wavelet transform technology is used to extract frequency features from each block of behavior data. Wavelet transform is a signal processing method that can analyze the characteristics of signals in time and frequency. In this case, wavelet transform is used to extract frequency features from the student's behavior data, which reflect the frequency and regularity of different behaviors of students in the learning process. For example, students may frequently access video content or participate in discussions in a specific time period. This behavior pattern can be quantified into a frequency vector through wavelet transform. With these frequency vectors, the system uses hidden Markov model to infer the state sequence of the target user. Hidden Markov model is a statistical model suitable for inferring hidden state sequences from observed data. Here, the observed data is the student's behavior frequency vector, and the hidden state is the student's learning state. Through this model, the system can infer the student's state in the learning process based on his behavior data, such as whether he is focused, whether he needs to review, whether he is interested in a certain knowledge point, etc. For example, if a student frequently accesses related resources when learning "Convolutional Neural Network (CNN)", the hidden Markov model may infer that he is learning this knowledge point in depth. Based on the inferred learning state sequence and knowledge point dependency chain, the system calculates the target user's learning position in the knowledge point dependency chain. The knowledge point dependency chain reflects the logical relationship between knowledge points, and the student's learning state sequence reflects his learning progress and mastery. Combining the two, the system can determine which knowledge point the student is currently learning. For example, if a student jumps directly to "Bayesian Network" after learning "Decision Tree", the system can judge from the knowledge point dependency chain that he may be trying to understand the Bayesian Network, but may ignore the previous knowledge points in it. Finally, the system performs future trend analysis on the learning position to predict the target user's access knowledge points in the knowledge point dependency chain. Future trend analysis is a prediction model that predicts the knowledge points that students may visit next based on their current learning position, online behavior logs, and learning status sequences. For example, if a student shows high interest and comprehension when learning "Support Vector Machine (SVM)", the system may predict that he will visit the relevant content of "kernel function" next, because this is a key concept of SVM. In this way, the system can obtain the learning trajectory of students based on the knowledge points visited, online behavior logs, and learning positions.The learning trajectory not only reflects the student's learning path, but also includes information about the student's mastery of knowledge points, learning progress, and adjustment of learning strategies during the learning process. For example, if a student shows a low mastery probability when learning "neural networks", the system can predict that he may encounter difficulties in the knowledge point of "deep learning" and suggest that he strengthen his understanding of the basics of neural networks. For example, suppose a student is studying an artificial intelligence course, and his online behavior log shows that he logs in frequently, but watches videos for a short time and posts in the discussion area less frequently when learning "machine learning basics". Through wavelet transform, the system extracts his behavior frequency vector and infers that he may not have a deep understanding of "machine learning basics" through the hidden Markov model. Based on the knowledge point dependency chain, the system predicts that he may jump directly to "decision tree" learning, while ignoring the prerequisite knowledge point of "linear algebra". Through future trend analysis, the system prompts him that he needs to review the relevant content of "linear algebra" to better understand "decision tree". Through this comprehensive analysis, the system can accurately infer students' learning trajectories in the knowledge point dependency chain and provide students with personalized learning suggestions. It also provides teachers with a basis for optimizing teaching content and strategies, thereby improving the efficiency and effectiveness of online learning.

[0034] In a specific embodiment, the predicting the target user's mastery probability of each knowledge point in the course outline based on the learning trajectory and the online behavior log includes: Extracting time series features from the learning trajectory to obtain time series features of the learning trajectory; Encoding the temporal features of the learning trajectory through a preset attention transformer network to obtain a semantic vector of the learning trajectory; Using the Bayesian probabilistic graphical model, the learning trajectory semantic vector and the online behavior log are used to perform probability reasoning on the target user at each knowledge point in the course outline to obtain a learning behavior probability distribution matrix; wherein the learning behavior probability distribution matrix includes the conditional probability and transition probability of knowledge point learning; Perform memory enhancement reasoning on each knowledge point in the course outline based on the learning behavior probability distribution matrix to obtain a dynamic prediction sequence for mastering the knowledge points; Based on the knowledge point mastery dynamic prediction sequence, each of the course outline knowledge points is estimated to have a mastery probability interval, and the mastery probability upper and lower bounds are obtained; wherein the mastery probability upper and lower bounds include a mastery knowledge lower bound and a mastery knowledge upper bound; Based on the upper and lower bounds of the mastery probability, the mastery probability of the target user at each knowledge point in the course outline is predicted.

[0035] Specifically, first, the system extracts temporal features from the learning trajectory. The learning trajectory contains the learning path and time sequence of the student on the knowledge point dependency chain. Temporal feature extraction aims to identify temporal regularities and trends from these paths, such as learning frequency, learning interval, and allocation of learning time. For example, if a student frequently accesses learning resources related to "decision tree" within a period of time, this may indicate that he is more focused on learning this knowledge point. Then, these temporal features are encoded through the preset attention Transformer network to obtain the learning trajectory semantic vector. The Transformer network is a deep learning model that is good at processing sequence data. Through the self-attention mechanism, it can capture the relationship between different time points in the learning trajectory. For example, when learning "deep learning", students may be more interested in "convolutional neural network (CNN)", and the Transformer network can recognize and encode this behavior pattern. Using the Bayesian probabilistic graph model, the system performs probabilistic reasoning on the learning trajectory semantic vector and online behavior log to obtain the learning behavior probability distribution matrix. This is a multi-dimensional matrix that contains the conditional probability and transition probability of each knowledge point. For example, conditional probability can represent the probability of a student's mastery of a certain knowledge point, while transition probability reflects the probability of a student's transition from one knowledge point to another. The Bayesian model infers the probability of a student's mastery of knowledge by observing his behavior pattern. For example, if a student shows a higher mastery probability when learning "Support Vector Machine (SVM)", then he may find it easier to understand "Kernel Function". Based on the learning behavior probability distribution matrix, the system performs memory-enhanced reasoning on each syllabus knowledge point to obtain a dynamic prediction sequence of knowledge point mastery. This means that the system not only considers the current learning state, but also combines the relationship between students' past learning behaviors and knowledge points to dynamically update the prediction of the mastery probability of each knowledge point. For example, if a student shows a lower mastery probability when learning "Neural Network", but he performs well in learning "Deep Learning", the system may infer that he has mastered the basic concepts of neural networks in other ways. Next, the system estimates the mastery probability interval for each syllabus knowledge point based on the knowledge point mastery dynamic prediction sequence to obtain the upper and lower bounds of the mastery probability. The upper and lower bounds of the mastery probability include the lower bound of mastery knowledge and the upper bound of mastery knowledge, which reflects the possible range of students' mastery of knowledge points. For example, if a student shows a high probability of mastering "Basic Machine Learning Fundamentals" but performs averagely on "Bayesian Networks", the system may estimate that the upper bound of his probability of mastering "Basic Machine Learning Fundamentals" is high, while the lower bound of his probability of mastering "Bayesian Networks" is low. Ultimately, based on the upper and lower bounds of the mastery probability, the system predicts the target user's mastery probability of each syllabus knowledge point. The prediction here is not just a single value, but a probability interval that includes uncertainty. In this way, the system can more accurately reflect students' learning outcomes.For example, if a student shows high frequency of interaction and high accuracy of homework submission when learning "decision tree", the system may predict that he has a high probability of mastering this knowledge point and is close to the upper bound of mastering knowledge. For example, suppose a student is studying an artificial intelligence course and his learning trajectory shows that he has visited the learning resources of "convolutional neural network (CNN)" many times, participated in discussions, and performed well in related tests. Through temporal feature extraction and encoding of Transformer network, the system identifies that he has a strong interest in and high understanding ability of "convolutional neural network". The Bayesian probabilistic graph model infers that he has a high probability of mastering "convolutional neural network" and predicts that he may perform well in "deep learning". Through memory-enhanced reasoning and mastery probability interval estimation, the system obtains that the student's mastery probability interval on "convolutional neural network" is 0.8 to 0.95, indicating that he has a high degree of mastery of this knowledge point. Through this comprehensive analysis and prediction, the system can provide a more accurate and dynamic evaluation framework to help students better understand their learning progress and adjust their learning strategies. It also provides teachers with a basis for optimizing teaching content and strategies to improve the efficiency and effectiveness of online learning.

[0036] In a specific embodiment, the memory enhancement reasoning is performed on each knowledge point in the course outline based on the learning behavior probability distribution matrix to obtain a knowledge point mastery dynamic prediction sequence, including: Through a memory-enhanced neural network, dynamically reconstructing each of the course outline knowledge points based on the learning behavior probability distribution matrix to obtain memory-reconstructed knowledge points; Performing probabilistic semantic deconstruction on the memory reconstruction knowledge points to obtain a latent semantic space of the knowledge points; Performing path weighted reasoning on the learning behavior probability based on the knowledge point latent semantic space to obtain a knowledge point relevance reasoning graph; Performing probability path inference on the knowledge point association reasoning graph to obtain a knowledge point mastery probability evolution sequence; Based on the probability evolution sequence of mastering the knowledge points, time series prediction is performed on the knowledge points in the course outline to obtain a dynamic prediction sequence of mastering the knowledge points.

[0037] Specifically, first, through the memory-enhanced neural network (MANN), the system dynamically reconstructs the memory of each course outline knowledge point based on the learning behavior probability distribution matrix to obtain the memory-reconstructed knowledge point. The memory-enhanced neural network is a neural network that combines memory mechanisms and can simulate human memory and reasoning processes. In this case, it uses students' past learning behavior data to dynamically update the understanding and mastery of each knowledge point. For example, if a student shows a high mastery probability when learning "decision tree", but then performs poorly on "random forest", MANN can dynamically adjust the estimate of the mastery probability of "decision tree" based on these behavior patterns. Next, the memory-reconstructed knowledge point is probabilistically semantically deconstructed to obtain the knowledge point latent semantic space. Probabilistic semantic deconstruction is to perform semantic analysis and deconstruction on each knowledge point to extract its potential meaning and relationship. In this way, the system can identify the concepts, principles, applications, etc. represented by each knowledge point in the student's learning process. For example, the knowledge point "deep learning" may contain sub-concepts such as "convolutional neural network" and "recurrent neural network". Through probabilistic semantic deconstruction, these sub-concepts can be mapped to a latent semantic space. Based on the latent semantic space of knowledge points, the system performs path-weighted reasoning on the probability of learning behavior and obtains a knowledge point relevance reasoning graph. Here, path-weighted reasoning is a reasoning method that considers the relationship between knowledge points. The system assigns weights to each knowledge point and its associated path by analyzing the dependency between knowledge points and the student's learning path. For example, if a student shows a high probability of mastering when learning "Support Vector Machine (SVM)", the path weight of knowledge points related to SVM, such as "kernel function", may be increased. The system obtains a knowledge point mastery probability evolution sequence by performing probabilistic path inference on the knowledge point relevance reasoning graph. Probabilistic path inference is to predict how a student's mastery of knowledge points evolves over time by analyzing the probability transfer between knowledge points on the path. For example, if a student shows a low probability of mastering when learning "Linear Algebra", but he performs well when learning "Neural Network", the system can infer that he may have made up for the lack of linear algebra through practice or other learning methods, and the probability of mastering increases over time. Finally, based on the knowledge point mastery probability evolution sequence, the system performs time series prediction on the syllabus knowledge points and obtains a knowledge point mastery dynamic prediction sequence. This is a dynamic prediction process that takes into account the student's past learning behavior and current learning status to predict the probability of mastering the knowledge points in the future. For example, if a student shows a high probability of mastering "Basics of Machine Learning", the system may predict that he will also perform well in "Deep Learning" because there is a logical relationship and knowledge dependency between the two knowledge points. For example, suppose a student is studying an artificial intelligence course, and his learning behavior probability distribution matrix shows that he has a high probability of mastering "Convolutional Neural Network (CNN)".Through memory-enhanced neural networks, the system dynamically memorizes and reconstructs the knowledge points of "CNN" and identifies that the student's understanding of "CNN" has reached a high level. Through probabilistic semantic deconstruction, the system maps "CNN" to a latent semantic space, which includes concepts such as "convolution operation" and "pooling layer". Based on these semantic spaces, the system constructs a correlation reasoning graph of knowledge points related to "CNN" through path weighted reasoning, such as the path weight from "CNN" to "object detection" is high. Through probabilistic path inference, the system predicts that the student may show a high mastery probability when learning "object detection" because he has a good understanding of "CNN". Finally, based on this mastery probability evolution sequence, the system predicts that the student will also show a high mastery probability when learning other knowledge points related to "CNN", such as "object recognition", thus forming a knowledge point mastery dynamic prediction sequence. Through this comprehensive analysis and prediction, the system can provide a more dynamic and accurate learning assessment to help students understand their learning progress and adjust their learning strategies, while providing teachers with a basis for optimizing teaching content and strategies to improve the efficiency and effectiveness of online learning.

[0038] In a specific embodiment, the weight adjustment of each knowledge point in the course outline based on the mastery probability to obtain the adjusted knowledge point weight includes: Based on the mastery probability, information entropy covariance analysis is performed on each knowledge point in the course outline to obtain a knowledge point complexity tensor; Based on the knowledge point complexity tensor, the course outline knowledge points are subjected to probabilistic semantic association deconstruction to obtain a knowledge point topological association graph; Based on a preset graph attention network, conditional probability reasoning is performed on the knowledge point topological association graph to obtain a probability distribution of knowledge point association; Based on the probability distribution of the knowledge point relevance, a priori probability distribution modeling is performed on the knowledge points in the course outline to obtain a priori distribution of the knowledge point weights; Adaptively nonlinearly mapping the prior distribution of the knowledge point weights to obtain a knowledge point dynamic weight evolution sequence; wherein the knowledge point dynamic weight evolution sequence includes an attenuation coefficient of the knowledge point weight and a learning evolution trajectory; The weight of each knowledge point in the course outline is adjusted based on the dynamic weight evolution sequence of the knowledge point to obtain the adjusted knowledge point weight.

[0039] Specifically, first, the system performs information entropy covariance analysis on each knowledge point in the syllabus based on the probability of mastery to obtain the knowledge point complexity tensor. Information entropy is a way to measure information uncertainty, while covariance analysis can reveal the relationship between variables. Here, the system calculates the complexity of each knowledge point by analyzing the student's mastery probability of each knowledge point. For example, if a student shows a low mastery probability when learning "Bayesian network", the complexity tensor of this knowledge point may show a high complexity, because understanding Bayesian network requires a strong foundation in probability theory and statistics. Next, based on the knowledge point complexity tensor, the system performs probabilistic semantic association deconstruction on the knowledge points in the syllabus to obtain a knowledge point topological association graph. Probabilistic semantic association deconstruction is to identify the logical connections and dependencies between knowledge points by analyzing the semantic and probabilistic relationships between them. For example, "decision tree" may have a strong correlation with "random forest" because random forest is an ensemble learning method based on multiple decision trees. This association deconstruction forms a topological graph that shows the relationship between knowledge points. Based on the preset graph attention network, the system performs conditional probability reasoning on the topological association graph of knowledge points to obtain the probability distribution of knowledge point association. The graph attention network is a deep learning model that can capture the relationship between nodes in a graph structure. In this case, the system uses the graph attention network to analyze the association between knowledge points and infer the probability distribution of each knowledge point in the entire knowledge system. For example, if students show a high probability of mastering when learning "neural networks", then the probability of association of knowledge points related to neural networks, such as "convolutional neural networks", may be high. Based on the probability distribution of knowledge point association, the system models the prior probability distribution of the knowledge points in the course outline to obtain the prior distribution of knowledge point weights. The prior probability distribution here is a statistical model that pre-sets the initial weight of each knowledge point based on the student's past learning behavior and the association between knowledge points. For example, if a student performs well when learning "Basics of Machine Learning", then the related knowledge points may be given a higher prior weight. Next, the prior distribution of knowledge point weights is adaptively nonlinearly mapped to obtain a sequence of dynamic weight evolution of knowledge points. Adaptive nonlinear mapping is a method of dynamically adjusting weights that takes into account the progress of student learning and changes over time. For example, if a student shows a low probability of mastery when learning "Support Vector Machine (SVM)", but then improves his understanding of SVM through practice or other learning methods, the weight of SVM may increase over time. The dynamic weight evolution sequence of knowledge points includes the decay coefficient of the knowledge point weight and the learning evolution trajectory, reflecting the dynamic changes of students in the learning process. Finally, based on the dynamic weight evolution sequence of knowledge points, the system adjusts the weight of each course outline knowledge point to obtain the adjusted knowledge point weight. This adjustment reflects the student's learning status and the degree of mastery of the knowledge point.For example, if a student shows a higher probability of mastery when learning "deep learning", the weight of knowledge points related to deep learning may be increased to reflect the student's learning progress. For example, suppose a student is studying an artificial intelligence course and his mastery probability shows that he has a deeper understanding of "decision tree" and a lower mastery probability of "random forest". Through information entropy covariance analysis, the system concludes that "random forest" is more complex than "decision tree". Then, through probabilistic semantic association deconstruction, the system finds that there is a close logical relationship between "decision tree" and "random forest". Using the graph attention network, the system infers that "random forest" has a higher correlation with "decision tree" in the probability distribution of knowledge point correlation. Based on this correlation, the system establishes a higher prior weight distribution for "random forest". Subsequently, through adaptive nonlinear mapping, the system observes that the student improves his understanding of "random forest" through practice and review, so the weight of "random forest" gradually increases as learning progresses. Finally, the system adjusts the weight of "random forest" according to the student's learning trajectory and mastery probability, and obtains an adjusted knowledge point weight that reflects the student's learning progress. Through this comprehensive analysis and weight adjustment, the system can better understand students' learning status, adjust teaching content and strategies, and provide personalized learning guidance, thereby improving the effectiveness and efficiency of online learning.

[0040] In a specific embodiment, the evaluating the online status of the target user based on the adjusted knowledge point weights and the mastery probability of the course outline knowledge points corresponding to the adjusted knowledge point weights includes: Through the preset Gaussian mixture model, the joint distribution of the adjusted knowledge point weights and mastery probabilities is estimated to obtain the joint distribution structure data; Randomly sampling the joint distribution structure data by using a Monte Carlo simulation method to obtain virtual sample data; Performing pattern recognition on the virtual sample data to mine hidden potential patterns in the virtual sample data to obtain a potential pattern set; Inputting the potential pattern set into a preset support vector machine model for classification to obtain classification results; wherein the classification results include excellent learning state, medium learning state and poor learning state; The online status of the target user is determined and evaluated based on the classification result.

[0041] Specifically, first, the system estimates the joint distribution of the adjusted knowledge point weights and mastery probabilities through a preset Gaussian mixture model (GMM) to obtain joint distribution structure data. The Gaussian mixture model is a statistical model that can capture multiple potential distributions in the data. In this case, the GMM is used to combine the adjusted knowledge point weights and the student's mastery probabilities of these knowledge points to form a joint distribution that reflects the student's learning status. For example, if a student shows a high mastery probability when learning "Convolutional Neural Network (CNN)" and the weight of this knowledge point is high, then in the GMM, this state may be regarded as a high-probability learning state. Next, the system randomly samples the joint distribution structure data through the Monte Carlo simulation method to obtain virtual sample data. Monte Carlo simulation is a statistical method that approximates the solution of complex problems through random sampling. Through this simulation, the system extracts a large number of virtual samples from the joint distribution, which represent the possible learning states of the students. For example, if a student shows fluctuations when learning "Deep Learning", the simulation may produce some samples showing a high mastery probability and others showing a low mastery probability. Pattern recognition is performed on these virtual sample data to mine the hidden potential patterns and obtain a potential pattern set. Pattern recognition here refers to identifying the regularity and trend in students' learning behavior. For example, the system may find that some students show periodic behavior in the learning process, such as doing a lot of review every weekend. These patterns can be identified and classified into a potential pattern set. These potential pattern sets are input into the preset support vector machine (SVM) model for classification to obtain classification results. Support vector machine is a powerful classification algorithm that can find an optimal hyperplane in high-dimensional space to divide data points into different categories. Here, the SVM model is used to classify students' learning status into excellent learning status, medium learning status and poor learning status. For example, if a student shows a high probability of mastery when learning "Basics of Machine Learning" and his learning behavior matches the pattern of excellent learning status, then SVM may classify him as an excellent learning status. Based on these classification results, the system determines and evaluates the online status of the target user. This means that the system can judge the student's current learning status and potential learning ability based on the student's learning behavior pattern, mastery probability and adjusted knowledge point weight. For example, if a student shows a low mastery probability when learning "neural networks", but his learning behavior pattern shows that he is working hard to learn and review related content, the system may evaluate him as a medium learning state and recommend that he continue to work hard to improve his mastery. For example, suppose a student is studying an artificial intelligence course, and his adjusted knowledge point weights show that he attaches great importance to "decision trees", while the mastery probability shows that he has a deeper understanding of "decision trees". Through the Gaussian mixture model, the system combines these data to obtain a joint distribution that reflects his learning state.Through Monte Carlo simulation, the system extracts virtual samples from this joint distribution, which show the possible learning status of students in different time periods. Through pattern recognition, the system finds that students spend time reviewing and practicing "decision trees" related content every week. After inputting these patterns into the support vector machine model, the system classifies the student's learning status as an excellent learning status because his learning behavior matches the pattern of an excellent learning status. Based on this classification result, the system evaluates that the student performed well in learning "decision trees", recommends that he continue to maintain this learning method, and provides him with suggestions or resources for further learning. Through this comprehensive evaluation method, the system is able to provide a dynamic, personalized and accurate learning status assessment to help students adjust their learning strategies and teachers optimize teaching content. It also provides educational institutions with an analytical basis for students' learning progress, thereby improving the effectiveness of online education and students' learning experience.

[0042] The above describes the online evaluation method for the artificial intelligence course for college students in the embodiment of the present invention. The following describes the online evaluation device for the artificial intelligence course for college students in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an online evaluation device for artificial intelligence courses for college students includes: A construction module 21 is used to obtain the syllabus knowledge points of the artificial intelligence course, and construct a knowledge point dependency chain for the syllabus knowledge points to obtain a knowledge point dependency chain; wherein the knowledge point dependency chain includes the logical relationship between the pre-dependent knowledge points and the post-associated knowledge points of each knowledge point; An inference module 22 is used to obtain an online behavior log of a target user, and infer a learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log through a preset hidden Markov model; Prediction module 23, used for predicting the mastery probability of the target user in each knowledge point of the course outline based on the learning trajectory and the online behavior log; An adjustment module 24, configured to adjust the weight of each knowledge point in the course outline based on the mastery probability to obtain an adjusted knowledge point weight; The evaluation module 25 is used to evaluate the online status of the target user based on the mastery probability corresponding to the knowledge points in the course outline and the weight of the adjusted knowledge points.

[0043] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0044] Reference Figure 3 The present invention also provides a computer device in an embodiment, wherein the internal structure of the computer device can be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0045] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0046] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0047] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0048] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0049] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An online evaluation method for college students' artificial intelligence courses, characterized in that: The following steps are involved: Acquire the syllabus knowledge points of the artificial intelligence course, and construct a knowledge point dependency chain for the syllabus knowledge points to obtain a knowledge point dependency chain; wherein the knowledge point dependency chain includes the logical relationship between the pre-dependent knowledge points and the post-associated knowledge points of each knowledge point; Obtaining the online behavior log of the target user, and inferring the learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log through a preset hidden Markov model; Based on the learning trajectory and the online behavior log, predict the probability of the target user mastering each knowledge point in the course outline; Adjust the weight of each knowledge point in the course outline based on the mastery probability to obtain an adjusted knowledge point weight; The online status of the target user is evaluated based on the mastery probability corresponding to the knowledge points in the course outline and the weight of the knowledge points is adjusted.

2. The online evaluation method for artificial intelligence courses for college students according to claim 1 is characterized in that: The step of constructing a knowledge point dependency chain for the course outline knowledge points to obtain a knowledge point dependency chain includes: Performing semantic analysis on the course outline knowledge points through a preset natural language processing technology to obtain a set of analyzed knowledge points; Performing relationship analysis on the knowledge points in the parsed knowledge point set to establish a dependency graph between the knowledge points; Vectorizing the dependency graph to map each knowledge point in the course outline into a high-dimensional space to obtain a vectorized representation of the knowledge point; Performing modular analysis on the vectorized representation of the knowledge points, identifying closely related subsets between knowledge points in each syllabus, and obtaining a knowledge point module dependency chain; By using a path importance evaluation method, the knowledge point module dependency chain is screened to obtain a screening key path; The graph convolutional network algorithm is used to model the transfer effect between the knowledge points in the critical screening path to obtain the knowledge point dependency chain.

3. The online evaluation method for artificial intelligence courses for college students according to claim 1 is characterized in that: The inferring the learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log by using a preset hidden Markov model includes: The behavior data in the online behavior log is processed by block processing at a preset fixed time interval to obtain block behavior data blocks; wherein the behavior data includes login time data, login number data, video viewing time data, discussion area posting data and homework submission data; Using wavelet transform technology, extracting frequency features from each of the divided behavior data blocks to obtain an extracted frequency vector; Using a hidden Markov model, based on the extracted frequency vector, a state sequence inference is performed on the target user to obtain a learning state sequence; Based on the learning state sequence and the knowledge point dependency chain, inferring the learning position of the target user in the knowledge point dependency chain; Perform future trend analysis on the learning position to predict the target user's accessed knowledge points in the knowledge point dependency chain, and obtain a learning trajectory based on the accessed knowledge points, online behavior logs and the learning position.

4. The online evaluation method for artificial intelligence courses for college students according to claim 1 is characterized in that: The predicting the target user's mastery probability of each course outline knowledge point based on the learning trajectory and the online behavior log includes: Extracting time series features from the learning trajectory to obtain time series features of the learning trajectory; Encoding the temporal features of the learning trajectory through a preset attention transformer network to obtain a semantic vector of the learning trajectory; Using the Bayesian probabilistic graphical model, the learning trajectory semantic vector and the online behavior log are used to perform probability reasoning on the target user at each knowledge point in the course outline to obtain a learning behavior probability distribution matrix; wherein the learning behavior probability distribution matrix includes the conditional probability and transition probability of knowledge point learning; Perform memory enhancement reasoning on each knowledge point in the course outline based on the learning behavior probability distribution matrix to obtain a dynamic prediction sequence for mastering the knowledge points; Based on the knowledge point mastery dynamic prediction sequence, each of the course outline knowledge points is estimated to have a mastery probability interval, and the mastery probability upper and lower bounds are obtained; wherein the mastery probability upper and lower bounds include a mastery knowledge lower bound and a mastery knowledge upper bound; Based on the upper and lower bounds of the mastery probability, the mastery probability of the target user at each knowledge point in the course outline is predicted.

5. The online evaluation method for artificial intelligence courses for college students according to claim 4 is characterized in that: The memory enhancement reasoning is performed on each knowledge point in the course outline based on the learning behavior probability distribution matrix to obtain a dynamic prediction sequence of knowledge point mastery, including: Through a memory-enhanced neural network, dynamically reconstructing each of the course outline knowledge points based on the learning behavior probability distribution matrix to obtain memory-reconstructed knowledge points; Performing probabilistic semantic deconstruction on the memory reconstruction knowledge points to obtain a latent semantic space of the knowledge points; Performing path weighted reasoning on the learning behavior probability based on the knowledge point latent semantic space to obtain a knowledge point relevance reasoning graph; Performing probability path inference on the knowledge point association reasoning graph to obtain a knowledge point mastery probability evolution sequence; Based on the probability evolution sequence of mastering the knowledge points, time series prediction is performed on the knowledge points in the course outline to obtain a dynamic prediction sequence of mastering the knowledge points.

6. The online evaluation method for artificial intelligence courses for college students according to claim 1 is characterized in that: The step of adjusting the weight of each knowledge point in the course outline based on the mastery probability to obtain the adjusted knowledge point weight includes: Based on the mastery probability, information entropy covariance analysis is performed on each knowledge point in the course outline to obtain a knowledge point complexity tensor; Based on the knowledge point complexity tensor, the course outline knowledge points are subjected to probabilistic semantic association deconstruction to obtain a knowledge point topological association graph; Based on a preset graph attention network, conditional probability reasoning is performed on the knowledge point topological association graph to obtain a probability distribution of knowledge point association; Based on the probability distribution of the knowledge point relevance, a priori probability distribution modeling is performed on the knowledge points in the course outline to obtain a priori distribution of the knowledge point weights; Adaptively nonlinearly mapping the prior distribution of the knowledge point weights to obtain a knowledge point dynamic weight evolution sequence; wherein the knowledge point dynamic weight evolution sequence includes an attenuation coefficient of the knowledge point weight and a learning evolution trajectory; The weight of each knowledge point in the course outline is adjusted based on the dynamic weight evolution sequence of the knowledge point to obtain the adjusted knowledge point weight.

7. The online evaluation method for artificial intelligence courses for college students according to claim 1 is characterized in that: The evaluating the online status of the target user based on the adjusted knowledge point weights and the mastery probability of the course outline knowledge points corresponding to the adjusted knowledge point weights includes: Through the preset Gaussian mixture model, the joint distribution of the adjusted knowledge point weights and mastery probabilities is estimated to obtain the joint distribution structure data; Randomly sampling the joint distribution structure data by using a Monte Carlo simulation method to obtain virtual sample data; Performing pattern recognition on the virtual sample data to mine hidden potential patterns in the virtual sample data to obtain a potential pattern set; Inputting the potential pattern set into a preset support vector machine model for classification to obtain classification results; wherein the classification results include excellent learning state, medium learning state and poor learning state; The online status of the target user is determined and evaluated based on the classification result.

8. An online evaluation device for artificial intelligence courses for college students, characterized in that: include: A construction module is used to obtain the syllabus knowledge points of the artificial intelligence course, and construct a knowledge point dependency chain for the syllabus knowledge points to obtain a knowledge point dependency chain; wherein the knowledge point dependency chain includes the logical relationship between the pre-dependent knowledge points and the post-associated knowledge points of each knowledge point; An inference module, used to obtain the online behavior log of the target user, and infer the learning trajectory of the target user in the knowledge point dependency chain based on the online behavior log through a preset hidden Markov model; A prediction module, used to predict the probability of the target user mastering each knowledge point in the course outline based on the learning trajectory and the online behavior log; An adjustment module, configured to adjust the weight of each knowledge point in the course outline based on the mastery probability to obtain an adjusted knowledge point weight; An evaluation module is used to evaluate the online status of the target user based on the mastery probability corresponding to the knowledge points in the course outline and the weight of the adjusted knowledge points.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images