Test question accurate pushing system based on learning behavior analysis

Abstract

The application discloses a test question accurate pushing system based on learning behavior analysis, and particularly relates to the technical field of test question accurate recommendation, and comprises a learning ability dynamic evaluation module, a multi-task model refinement module, a speculation completion module and a self-adaptive recommendation module.The learning ability dynamic evaluation module constructs a time series model according to user learning behavior data to predict the future learning ability of the user for one-time classification.The multi-task model refinement module refines the result of one-time classification.The speculation completion module generates learning behavior features based on a graph neural network in combination with known learning behavior data.The self-adaptive recommendation module performs personalized recommendation on different types of users according to the output result of the speculation completion module.The learning ability of the user is evaluated by using the time series model, the classification is refined in combination with the multi-task learning model, and personalized test question recommendation is realized by using the graph neural network to speculate potential learning behavior, so that the data sparsity problem is effectively solved, and the recommendation accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, and more specifically, to a test question delivery system based on learning behavior analysis. Background Technology

[0002] The potential targets for targeted test questions are usually arranged in a normal distribution. Students with extremely strong or weak learning abilities are only a very small minority. The needs of this small number of users present the problem of data sparsity, making it difficult to carry out precise and personalized push notifications.

[0003] To address the issue of data sparsity, a common approach is to use a cold start approach, which involves trial processing based on the personal experience and recommendation rules of industry experts. This method is highly subjective and requires a wide range of rules, making it difficult to adapt to a small number of users with extreme data types. Another common approach is to extract and transfer data from other domains to the target domain based on transfer learning. However, there are differences between the source and target domains, and smoothing these differences and mastering the behavioral logic of the learned data pose significant challenges in practical applications.

[0004] To address the aforementioned shortcomings, a technical solution is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a test question delivery system based on learning behavior analysis to address the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a test question accurate push system based on learning behavior analysis, including a learning ability dynamic assessment module, a multi-task model refinement module, a prediction completion module, and an adaptive recommendation module;

[0007] The dynamic learning ability assessment module is used to build a time series model based on the user's learning behavior data to predict the user's future learning ability, and classify the user's learning status based on the user's future learning ability and the user's current learning performance.

[0008] The multi-task model refinement module is used to refine the results of a single classification. After labeling different questions, a multi-task learning model is established. The multi-task learning model includes a first output layer for predicting the questions requested by the user, a second output layer for classifying the question type, and a third output layer for evaluating the difficulty of the questions. Based on the output results of the multi-task shared underlying network, target users with extremely strong learning ability and extremely weak learning ability are labeled.

[0009] The inference completion module is used to infer the potential learning behavior of the target user using a graph neural network-based method, and to generate learning behavior features by combining known learning behavior data.

[0010] The adaptive recommendation module is used to provide personalized recommendations for different types of users based on the output of the inference completion module.

[0011] Preferably, a feature vector representing changes in a user's learning ability is constructed based on features extracted from time series data. The method for constructing the feature vector is as follows:

[0012] The number of questions answered within the calibration time window t is N. t The correct answer rate within the time window t is A. t A t The calculation expression is as follows In the formula, C t Let T be the number of correct answers within a time window t, and T be the average answering time. t The calculation expression is: Among them, t i The time allotted for each question is F, and the frequency of reviewing incorrect answers is F. t The calculation expression is: Learning progress rate In the formula, δ and ∈ are both extremely small positive numbers, A t-1 The correct answer rate in the previous time window is given by S, and the learning stability is given by S. t The calculation expression is: In the formula, n is the amount of data related to academic performance, and x i For the i-th learning score, the feature vector v representing the change in the user's learning ability is... t The form is v t =[N t A t T t F t P t S t ].

[0013] Preferably, the feature vector v t After inputting into a time series model, the output is a prediction of the user's future learning ability. A logistic regression model is then used to classify users based on these predictions. The specific method is as follows:

[0014] Establish a logistic regression model In the formula, P(y=1|X) is the user's classification bias value, y is a binary dependent variable representing the judgment category of the user's learning ability, x is a vector of independent variables including the output prediction of the time series model, the user's learning performance, and various features, β0, β1, ... β m Here are the coefficients of the logistic regression model, where m is the number of independent variable vectors and m is a positive integer;

[0015] Set the first threshold P for classification.f And the second threshold P for the first classification s And the first threshold P for classification f Less than the second threshold P for classification s When the calculated user classification bias value P is less than or equal to the first classification threshold P f If the calculated classification deviation value P of the user is greater than or equal to the second threshold P of the first classification, then the user is labeled as a Class A target user. s If the calculated classification deviation value P of the user is greater than the first classification threshold P, then the user is labeled as a Class B target user. f And less than the second threshold P for the first classification s At that time, the user is marked as a non-target user.

[0016] Preferably, the logic for establishing a multi-task learning model is as follows:

[0017] Let the multi-task learning model be Z = f(X). u X q In the formula, Z is the shared feature vector, and X is the shared feature vector. u It is a user characteristic, X q Here, f represents the question features, f is the underlying shared network, and the first output layer of the multi-task learning model is used to predict the user-requested question. The loss function is the cross-entropy function, expressed as follows: In the formula, y j For target classification, p j Let B be the predicted category probability, and B be the number of questions. The second output layer of the multi-task learning model is used to classify questions based on their labels, determining their question type. The loss function is the cross-entropy function, expressed as follows: In the formula, E k As the true label for the question type, q k Let V be the probability of the predicted question type, and V be the number of question types. The third output layer of the multi-task learning model is used to predict the difficulty of the questions, outputting the difficulty value of the question. The loss function is the mean squared error loss, expressed as follows: In the formula, D h This represents the true difficulty level of the question. The overall loss function is a weighted sum of L1, L2, and L3, representing the difficulty value for prediction.

[0018] Preferably, the logic for refining the classification based on the user's learning ability prediction results using a multi-task learning model, building upon a single classification, is as follows:

[0019] Define the dominant ability function S(X) u The output of the underlying network of the multi-task learning model is represented as S(X). u )=f cl (Z u In the formula, Zu f is a user feature extracted from a shared network. cl Let E be the classification function, and set the threshold for dominant ability classification as E. T Then the output type in Among them, T y =1 indicates that the user has extremely strong learning ability, T y =0 indicates that the user's learning ability is extremely weak. For target users of type A, when the output type is T y When = 0, the user is marked as a user with extremely weak dominant ability. When the output type is T y When = 1, the user is marked as an invisible floating user;

[0020] For Class B users, when the output type is T y =1, marking the user as a user with extremely strong explicit ability, when the user's output type is T. y When the value is 0, the user is marked as an invisible floating user.

[0021] Preferably, the logic for generating learning behavior features based on graph neural networks is as follows:

[0022] The graph's nodes include user nodes, question nodes, and knowledge point nodes. User nodes represent individual users, question nodes represent questions, and knowledge point nodes represent the knowledge points involved in the questions. The graph's edges include user-question edges and question-knowledge point edges. User-question edges represent relationships where a user has answered a question, and their weights are assigned based on the user's accuracy and answering time. Question-knowledge point edges represent knowledge points associated with a question, and their weights are assigned based on the question's difficulty and the importance of the knowledge point. The graph is represented as G = (W, A), where W is the node set, including users, questions, and knowledge points, and A is the edge set, including the relationships between users and questions, and between questions and knowledge points. Information is propagated through connections between nodes, updating the feature representation of each node, and generating node features using known learning behavior data.

[0023] Preferably, the logic for recommending test questions based on the output of the graph neural network is as follows:

[0024] By integrating user learning behavior data with the relationships between questions and knowledge points through convolutional processing of a graph neural network, the graph neural network is trained to obtain the user's feature representation as h. u The expression is h u =GNN(X u In the formula, X u Given the user's feature input, GNN stands for Graph Neural Network, h uThe graph convolution is used to represent user features. For users with extremely strong explicit abilities, a regression model is used to predict their future learning ability and recommend test questions. For users with extremely weak explicit abilities, a classification model is used to predict their learned knowledge and recommend test questions.

[0025] The preferred method for predicting users' future learning ability performance and recommending test questions using a regression model is as follows:

[0026] Define the regression model as To predict the learning abilities of users with exceptionally strong explicit skills, the difficulty of recommended questions is dynamically adjusted based on the user's learning ability performance predicted by the regression model. The adjustment logic is as follows: In the formula, To recommend the difficulty level of the test questions, A function to adjust the difficulty of questions based on the user's predicted learning performance.

[0027] The preferred method for using a classification model to predict a user's learned knowledge and recommend test questions is as follows:

[0028] Define the classification model as P(y|h) u ) = softmax(W2·σ(W1·h) u +b1)+b2), where W1 and W2 are the weight matrices of the classification model, σ is the activation function, and the output is processed by the softmax function to obtain the user's learned category distribution, and the user's features h are then processed. u and the characteristics of the question h q Matching is performed, and the similarity between the matches is calculated to recommend test questions. The method for calculating the similarity between matches is as follows: The higher the similarity of the match, the stronger the recommendation priority.

[0029] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0030] This application addresses the data sparsity problem of extreme users by employing a dual classification of user characteristics. It includes dynamic assessment of learning ability, multi-task model refinement, inference completion modules, and adaptive recommendation modules. It constructs a time-series model using learning behavior data to predict users' future learning ability, refines user classification using a multi-task learning model, infers potential learning behaviors using graph neural networks, generates personalized recommendations, assesses learning ability in real time, provides accurate test item recommendations, improves assessment accuracy through multi-task learning, avoids redundant training, and utilizes graph neural networks to capture complex learning behavior relationships, thereby enhancing the quality of test item recommendations. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0032] Figure 1 This is a system module diagram of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0034] Example 1: Please refer to Figure 1 As shown, the present invention is a test question precise push system based on learning behavior analysis, including a learning ability dynamic assessment module, a multi-task model refinement module, a prediction completion module, and an adaptive recommendation module;

[0035] The dynamic learning ability assessment module is used to build a time series model based on the user's learning behavior data to predict the user's future learning ability, and classify the user's learning status based on the user's future learning ability and the user's current learning performance.

[0036] The multi-task model refinement module is used to refine the results of a single classification. After labeling different questions, a multi-task learning model is established. The multi-task learning model includes a first output layer for predicting the questions requested by the user, a second output layer for classifying the question type, and a third output layer for evaluating the difficulty of the questions. Based on the output results of the multi-task shared underlying network, target users with extremely strong learning ability and extremely weak learning ability are labeled.

[0037] The inference completion module is used to infer the potential learning behavior of the target user using a graph neural network-based method, and to generate learning behavior features by combining known learning behavior data.

[0038] The adaptive recommendation module is used to provide personalized recommendations for different types of users based on the output of the inference completion module.

[0039] The dynamic learning ability assessment module is used to build a time series model based on the user's learning behavior data to predict the user's future learning ability, and classify the user's learning status based on the user's future learning ability and the user's current learning performance.

[0040] User learning behavior data includes the number of questions answered, accuracy rate, answering time, and frequency of reviewing incorrect questions. A timestamp is added to each user's learning behavior, transforming the data into time series data. A time window is defined, representing a period of learning time. The changing trends of the learning behavior data within each window are analyzed. Based on the time series data, features are extracted to construct a feature vector representing changes in the user's learning ability. The method for constructing the feature vector is as follows:

[0041] The number of questions answered within the calibration time window t is N. t The correct answer rate within the time window t is A. t A t The calculation expression is as follows In the formula, C t Let T be the number of correct answers within a time window t, and T be the average answering time. t The calculation expression is: Among them, t i The time allotted for each question is F, and the frequency of reviewing incorrect answers is F. t The calculation expression is: Learning progress rate In the formula, δ and ∈ are both extremely small positive numbers, A t-1 The correct answer rate in the previous time window is given by S, and the learning stability is given by S. t The calculation expression is: In the formula, n is the amount of data related to academic performance, and x i For the i-th learning score, the feature vector v representing the change in the user's learning ability is... t The form is v t =[N t A t T t F t P t S t ].

[0042] The feature vector v t After inputting into a time series model, the output is a prediction of the user's future learning ability. A logistic regression model is then used to classify users based on these predictions. The specific method is as follows:

[0043] Establish a logistic regression model In the formula, P(y=1|X) is the user's classification bias value, y is a binary dependent variable representing the judgment category of the user's learning ability, x is a vector of independent variables including the output prediction of the time series model, the user's learning performance, and various features, β0, β1, ... β m Here are the coefficients of the logistic regression model, where m is the number of independent variable vectors and m is a positive integer;

[0044] Set the first threshold P for classification. f And the second threshold P for the first classification s And the first threshold P for classification f Less than the second threshold P for classification s When the calculated user classification bias value P is less than or equal to the first classification threshold P f If the calculated classification deviation value P of the user is greater than or equal to the second threshold P of the first classification, then the user is labeled as a Class A target user. s If the calculated classification deviation value P of the user is greater than the first classification threshold P, then the user is labeled as a Class B target user. f And less than the second threshold P for the first classification s At that time, the user is marked as a non-target user.

[0045] The target users of type A have low learning ability and consistently poor learning performance, while the target users of type B have high learning ability and consistently good learning performance. Both types of target users belong to extreme user types with data sparsity problems. The existing data may be insufficient to meet the recommendation needs of extreme user types. The fundamental reason is that the extremely small number of extreme user types has the limitation of data sparsity.

[0046] Traditional recommendation algorithm layout logic uses cold start processing or transfer learning to solve the problem. However, in test question recommendation, the interaction of test question recommendation and evaluation is always present and inevitable. Users who do not have the need for test question feedback will not continuously take test questions. The need definitely exists, but the number of extreme users is too small to be analyzed with a sufficient amount of learning behavior data. On the other hand, the transfer learning-based solution has high requirements for the source domain and target domain of the transfer learning application, which is difficult to implement in practice.

[0047] Time series analysis can capture the trends and periodicity of learning behavior over time, revealing the inherent laws in the learning process. Based on the results of time series analysis, personalized learning paths and test recommendations can be generated. Real-time assessment of changes in learning ability helps to adjust the test recommendation content in a timely manner. On the other hand, the learning ability of different users may not always remain constant. There is a possibility that some users' learning ability performance will change from a low level to a high level or from a high level to a low level. Therefore, real-time assessment of users is necessary, and thus, users need to be classified in two ways.

[0048] The multi-task model refinement module is used to refine the results of a single classification. After labeling different questions, a multi-task learning model is established. The multi-task learning model includes a first output layer for predicting the questions requested by the user, a second output layer for classifying the question type, and a third output layer for evaluating the difficulty of the questions. Based on the output results of the multi-task shared underlying network, target users with extremely strong learning ability and extremely weak learning ability are labeled.

[0049] The multi-task learning model refines the assessment of different users' learning abilities and labels the questions. The model should include three tasks: predicting the questions users need, classifying the question type, and assessing the difficulty of the questions. Each task shares the feature representations of the underlying network and is trained together through a multi-task learning framework. The underlying network of the multi-task learning model includes three input layers:

[0050] User characteristic input: User learning behavior data, including accuracy, answering speed, and learned knowledge points;

[0051] Question feature input: Question tag information, including question type, related knowledge point, and question difficulty;

[0052] Historical learning paths: The user's learning path sequence is used as a time series input;

[0053] The logic for establishing a multi-task learning model is as follows:

[0054] Let the multi-task learning model be Z = f(X). u X q In the formula, Z is the shared feature vector, and X is the shared feature vector. u It is a user characteristic, X q Here, f represents the question features, f is the underlying shared network, and the first output layer of the multi-task learning model is used to predict the user-requested question. The loss function is the cross-entropy function, expressed as follows: In the formula, y j For target classification, p j Let B be the predicted category probability, and B be the number of questions. The second output layer of the multi-task learning model is used to classify questions based on their labels, determining their question type. The loss function is the cross-entropy function, expressed as follows: In the formula, E k As the true label for the question type, q k Let V be the probability of the predicted question type, and V be the number of question types. The third output layer of the multi-task learning model is used to predict the difficulty of the questions, outputting the difficulty value of the question. The loss function is the mean squared error loss, expressed as follows: In the formula, D h This represents the true difficulty level of the question. The overall loss function is a weighted sum of L1, L2, and L3, representing the difficulty value to be predicted.

[0055] Based on the prediction results of users' learning abilities using a multi-task learning model, the target users of category A and category B obtained in the first classification are filtered out. Users with extremely strong and extremely weak learning abilities are further refined into categories. The logic for this refinement is as follows:

[0056] Define the dominant ability function S(X) u The underlying network output representation based on the multi-task learning model, S(X) u )=f cl (Z u In the formula, Z u f is a user feature extracted from a shared network. cl Let E be the classification function, and set the threshold for dominant ability classification as E. T Then the output type in, Among them, T y =1 indicates that the user has extremely strong learning ability, T y =0 indicates that the user's learning ability is extremely weak. For target users of type A, when the output type is T y When = 0, the user is marked as a user with extremely weak dominant ability. When the output type is T y When = 1, the user is marked as an invisible floating user;

[0057] For Class B users, when the output type is T y =1, marking the user as a user with extremely strong explicit ability, when the user's output type is T. y When the value is 0, the user is marked as an invisible floating user.

[0058] It should be noted that the weights of each loss function are determined by professionals in the field through hyperparameter tuning.

[0059] Multi-task learning models are trained specifically for different question types and difficulty levels, making learning more accurate and efficient. Compared to traditional single-task learning models, multi-task learning avoids repetitive training, saving time and resources. It evaluates users' learning abilities from multiple dimensions, including their mastery of different question types, problem-solving speed, and accuracy. This multi-dimensional evaluation provides a more comprehensive understanding of the user's learning status. By sharing representation layers between different tasks, multi-task learning enables the model to learn more general and robust feature representations. Meanwhile, auxiliary tasks have a regularization effect on the main task, reducing overfitting and improving the model's stability and reliability.

[0060] The inference completion module is used to infer the potential learning behavior of the target user using a graph neural network-based method, and to generate learning behavior features by combining known learning behavior data.

[0061] A learning behavior graph is constructed based on a graph neural network. This graph connects users with extremely strong and extremely weak explicit abilities to questions and knowledge point entities, thereby abstracting learning behavior into a graph model. The construction logic of the graph neural network is as follows:

[0062] The graph's nodes include user nodes, question nodes, and knowledge point nodes. User nodes represent individual users, question nodes represent questions, and knowledge point nodes represent the knowledge points involved in the questions. The graph's edges include user-question edges and question-knowledge point edges. User-question edges represent relationships where a user has answered a question, and their weights are assigned based on the user's accuracy and answering time. Question-knowledge point edges represent knowledge points associated with a question, and their weights are assigned based on the question's difficulty and the importance of the knowledge point. The graph is represented as G = (W, A), where W is the node set, including users, questions, and knowledge points, and A is the edge set, including the relationships between users and questions, and between questions and knowledge points. Information is propagated through connections between nodes, updating the feature representation of each node, and generating node features using known learning behavior data.

[0063] The adaptive recommendation module is used to provide personalized recommendations for different types of users based on the output of the inference completion module.

[0064] By integrating user learning behavior data with the relationships between questions and knowledge points through convolutional processing of a graph neural network, the graph neural network is trained to obtain the user's feature representation as h. u The expression is h u =GNN(X u In the formula, X u Given the user's feature input, GNN stands for Graph Neural Network, h u The user feature representation after graph convolution is used. For users with extremely strong explicit abilities, a regression model is used to predict the user's future learning ability and recommend test questions. For users with extremely weak explicit abilities, a classification model is used to predict the user's learned knowledge and recommend test questions.

[0065] The method of using regression models to predict users' future learning ability and recommend test questions is as follows:

[0066] Define the regression model as To predict the learning abilities of users with exceptionally strong explicit skills, the difficulty of recommended questions is dynamically adjusted based on the user's learning ability performance predicted by the regression model. The adjustment logic is as follows: In the formula, To recommend the difficulty level of the test questions, A function to adjust the difficulty of questions based on the user's predicted learning performance;

[0067] The method of using a classification model to predict a user's learned knowledge and recommend test questions is as follows:

[0068] Define the classification model as P(y|h) u ) = softmax(W2·σ(W1·h) u +b1)+b2), where W1 and W2 are the weight matrices of the classification model, σ is the activation function, and the output is processed by the softmax function to obtain the user's learned category distribution, and the user's features h are then processed. u and the characteristics of the question h q Matching is performed, and the similarity between the matches is calculated to recommend test questions. The method for calculating the similarity between matches is as follows: The higher the similarity of the match, the stronger the recommendation priority.

[0069] This application addresses the data sparsity problem of extreme users by employing a dual classification of user characteristics. It includes dynamic assessment of learning ability, multi-task model refinement, inference completion modules, and adaptive recommendation modules. It constructs a time-series model using learning behavior data to predict users' future learning ability, refines user classification using a multi-task learning model, infers potential learning behaviors using graph neural networks, generates personalized recommendations, assesses learning ability in real time, provides accurate test item recommendations, improves assessment accuracy through multi-task learning, avoids redundant training, and utilizes graph neural networks to capture complex learning behavior relationships, thereby enhancing the quality of test item recommendations.

[0070] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0071] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as computer program goods. The computer program goods include one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0072] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0073] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0074] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0075] If the aforementioned functions are implemented as software functional units and sold or used as independent goods, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of software goods. This computer software goods are stored in a storage medium and include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0076] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A test question delivery system based on learning behavior analysis, characterized in that: It includes a dynamic learning ability assessment module, a multi-task model refinement module, a speculative completion module, and an adaptive recommendation module; The dynamic learning ability assessment module is used to build a time series model based on the user's learning behavior data to predict the user's future learning ability, and classify the user's learning status based on the user's future learning ability and the user's current learning performance. The multi-task model refinement module is used to refine the results of a single classification. After labeling different questions, a multi-task learning model is established. The multi-task learning model includes a first output layer for predicting the questions requested by the user, a second output layer for classifying the question type, and a third output layer for evaluating the difficulty of the questions. Based on the output results of the multi-task shared underlying network, target users with extremely strong learning ability and extremely weak learning ability are labeled. The inference completion module is used to infer the potential learning behavior of the target user using a graph neural network-based method, and to generate learning behavior features by combining known learning behavior data. The adaptive recommendation module is used to make personalized recommendations for different types of users based on the output of the inference completion module; Based on the prediction results of users' learning abilities using a multi-task learning model, the logic for refining the classification on the basis of a single classification is as follows: Define the dominant ability function The output of the underlying network based on the multi-task learning model is represented as In the formula, For user features extracted from a shared network, Let be the classification function, and set the threshold for dominant ability classification as follows: Then the output type ,in ,in, This indicates that the user has extremely strong learning abilities. This indicates that the user's learning ability is extremely weak. For target users of type A, when the output type is... When the user is marked as having extremely weak explicit capabilities, and the output type is... At that time, mark the user as an invisible floating user; For target users of type B, when the output type is Mark users as having extremely strong explicit capabilities when the output type is At that time, mark the user as an invisible floating user; Category A target users have lower learning abilities and consistently poor learning performance, while Category B target users have stronger learning abilities and consistently better learning performance. The logic for generating learning behavior features based on graph neural networks is as follows: The graph's nodes include user nodes, question nodes, and knowledge point nodes. User nodes represent individual users, question nodes represent questions, and knowledge point nodes represent the knowledge points involved in the questions. The graph's edges include user-question edges and question-knowledge point edges. User-question edges represent relationships where a user has answered a question; the weight of a user-question edge is assigned based on the user's accuracy and answering time. Question-knowledge point edges represent knowledge points associated with a question; the weight of a question-knowledge point edge is assigned based on the question's difficulty and the importance of the knowledge point. The graph is represented as follows: ,in, The graph is a set of nodes, which includes users, questions, and knowledge points. The edge set of the graph includes the association between users and questions, as well as the association between questions and knowledge points. Information is propagated through the connections between nodes, the feature representation of each node is updated, and node features are generated using known learning behavior data. The logic for recommending test questions based on the output of the graph neural network is as follows: By integrating user learning behavior data with the relationships between questions and knowledge points through convolutional processing of a graph neural network, the graph neural network is trained to obtain the user's feature representation. The expression is In the formula, For user feature input, This represents a graph neural network. The graph convolution is used to represent user features. For users with extremely strong explicit abilities, a regression model is used to predict their future learning ability and recommend test questions. For users with extremely weak explicit abilities, a classification model is used to predict their learned knowledge and recommend test questions.

2. The test question precise delivery system based on learning behavior analysis according to claim 1, characterized in that, Based on time series data, a feature vector representing changes in a user's learning ability is constructed. The method for constructing the feature vector is as follows: The number of answers within the calibrated time window t is The correct answer rate within the time window t is , The calculation expression is as follows In the formula, The number of correct answers within the time window t, and the average answering time are... The calculation expression is: ,in, The time allotted for each question, and the frequency of reviewing incorrect answers are as follows: The calculation expression is: Learning progress rate In the formula, and All are extremely small positive numbers. The learning stability is the accuracy rate of answering questions in the previous time window. The calculation expression is: In the formula, n represents the amount of data related to academic performance. For the i-th learning score, the feature vector representing the change in the user's learning ability is... The form is .

3. The test question precise delivery system based on learning behavior analysis according to claim 2, characterized in that, eigenvectors After inputting into a time series model, the output is a prediction of the user's future learning ability. A logistic regression model is then used to classify users based on these predictions. The specific method is as follows: Establish a logistic regression model In the formula, Let y be the user's classification bias value, y be a binary dependent variable representing the judgment category of the user's learning ability, and x be a vector of independent variables, including the output prediction of the time series model, the user's learning performance, and various features. Here are the coefficients of the logistic regression model, where m is the number of independent variable vectors and m is a positive integer; Set the first threshold for classification and the second threshold for classification And the first threshold for classification Less than the second threshold of the first classification When the calculated user classification bias value Less than or equal to the first threshold for classification When the user is identified as a Class A target user, the calculated classification deviation value is used. Greater than or equal to the second threshold of the first classification When the user is identified as a Class B target user, the calculated classification deviation value is used to determine the classification deviation. Greater than the first threshold of classification And less than the second threshold of the first classification At that time, the user is marked as a non-target user.

4. The test question precise delivery system based on learning behavior analysis according to claim 1, characterized in that, The logic for establishing a multi-task learning model is as follows: Set the multi-task learning model as In the formula, Z is the shared feature vector. It is a user characteristic. These are characteristics of the question. It is a low-level shared network. The first output layer of the multi-task learning model is used to predict the user-requested question. The loss function is the cross-entropy function, expressed as follows: In the formula, Classify the target. Let B be the predicted category probability, and B be the number of questions. The second output layer of the multi-task learning model is used to classify questions based on their labels, determining their question type. The loss function is the cross-entropy function, expressed as follows: In the formula, As a true label for the question type, Let V be the probability of the predicted question type, and V be the number of question types. The third output layer of the multi-task learning model is used to predict the difficulty of the questions, outputting the difficulty value of the question. The loss function is the mean squared error loss, expressed as follows: In the formula, This represents the true difficulty level of the question. Let be the difficulty value for prediction, and the overall loss function be... , and The weighted sum.

5. The test question precise delivery system based on learning behavior analysis according to claim 1, characterized in that, The method of using regression models to predict users' future learning ability and recommend test questions is as follows: Define the regression model as , To predict the learning abilities of users with exceptionally strong explicit skills, the difficulty of recommended questions is dynamically adjusted based on the user's learning ability performance predicted by the regression model. The adjustment logic is as follows: In the formula, To recommend the difficulty level of the test questions, A function to adjust the difficulty of questions based on the user's predicted learning performance.

6. The test question precise delivery system based on learning behavior analysis according to claim 5, characterized in that, The method of using a classification model to predict a user's learned knowledge and recommend test questions is as follows: Define the classification model as In the formula, , This is the weight matrix of the classification model. As the activation function, the output is... The function obtains the distribution of categories that the user has learned and incorporates the user's features. and the characteristics of the question Matching is performed, and the similarity between the matches is calculated to recommend test questions. The method for calculating the similarity between matches is as follows: The higher the similarity of the match, the stronger the recommendation priority.

Images