A knowledge point intelligent navigation method for a learner competency model

Abstract

The application provides a knowledge point intelligent navigation method for a learner competence model, based on a low learning cost deep neural network model, from a knowledge point level, initial competence information of a learner and target knowledge points to be learned to reach a target competence are acquired, a current learnable knowledge point is recommended for the learner, the learner enters a course containing the knowledge point to learn, tests are conducted before and after learning, rewards obtained by performing a current action are calculated according to a reward function, and after learning is completed, the next knowledge point is recommended for the learner according to the obtained rewards, and the process is repeated until the learner learns the target knowledge point, reaches the target competence, and obtains a course path to be learned by the learner to reach the target competence, through integration and analysis of teaching data, a more accurate and personalized learning path is provided for students, and learning efficiency and effect of the students are improved.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, specifically relating to an intelligent navigation method for knowledge points based on learner competency models. Background Technology

[0002] Building a networked, digitalized, personalized, and lifelong education system is a crucial condition for realizing the construction of a learning society. Supported by numerous new technologies such as mobile internet, big data, artificial intelligence, and cloud computing, educational models, teaching methods, and learning styles are undergoing profound changes, propelling my country's education towards intelligence and smart technology. Among these advancements, the concepts of "focusing on individual learners' differences" and "providing appropriate education for each learner" are gaining increasing social consensus. Personalized education and personalized learning have become major pathways to addressing the current imbalance between supply and demand in education. In the new media environment, exemplified by social media, learning resources are abundant. How to utilize intelligent technology to provide learners with customized learning paths is a current hot topic in personalized learning research.

[0003] Existing personalized course recommendations are based on learners' previous learning levels and goals, with platforms customizing learning paths to ensure effective learning. Some personalized learning platforms also use machine learning algorithms to analyze learners' learning behaviors and abilities to recommend courses. However, this method also has some drawbacks. For example, the algorithm may favor recommending content that learners were previously interested in, relying too heavily on historical behavior, which could prevent learners from accessing newer knowledge. Furthermore, the algorithm's recommendations may be influenced by system biases, leading to skewed analyses of some learners and affecting the accuracy and fairness of the recommendation system. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides an intelligent knowledge point navigation method based on learner competency models. Based on the differences in learners' learning abilities, knowledge backgrounds, learning interests, and goals, it uses intelligent technology to customize a learning path that conforms to educational principles and achieves learning objectives for each learner, while simultaneously detecting the learner's learning status.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: a knowledge point intelligent navigation method oriented towards learner competency models. Based on a low-learning-cost deep neural network model, it obtains a learner's initial competency information and the target knowledge point to be learned to achieve the target competency at the knowledge point level. It recommends a currently learnable knowledge point to the learner, who then enters a course containing that knowledge point for learning. Tests are conducted before and after learning, and the reward obtained for completing the current action is calculated based on a reward function. After learning, the next knowledge point is recommended to the learner based on the obtained reward, and this process is repeated until the learner learns the target knowledge point and achieves the target competency, thus obtaining the course path that the learner needs to learn to achieve the target competency. Low-learning-cost deep neural network models transform the entire process into a state-based process. ,action ,award The Markov Decision Process (MDP) is composed of elements. By exploring how the MDP learns to generate a course path to achieve the target competency, the current state serves as the input of the agent. The agent outputs an action and receives a reward for performing this action in the current state to stimulate the agent to learn continuously. Among them, the knowledge points that can be learned at present are obtained from the candidate knowledge point set, and the candidate knowledge point set is obtained through the knowledge point navigation method.

[0006] Furthermore, specific methods for training deep neural network models with low learning costs include the following steps: S1, randomly initializes the parameters of a deep neural network model with low learning cost; S2, Initialize the experience pool H; S3, Initialize the new course set This collection is used to store courses that a learner has already studied and that have improved their competence. Initially empty; S4, randomly select a learner And acquire its target knowledge points ; S5, Initialize knowledge point learning records and obtain the initial state. ; S6, obtain a set of candidate knowledge points through the knowledge point navigation method; S7: Select a knowledge point from the candidate knowledge point set as the current action. Then proceed to the course containing that knowledge point. To learn; S8 will complete the action. The newly acquired competence is compared with the initial competence to obtain the corresponding reward for the current state. Update the knowledge point learning record L and obtain the status of the next step. ; S9, select the course at this time Put in In the set; S10, will Place it into experience pool H.

[0007] S11, using state replace ; S12, Randomly collect a sample from the experience replay pool H. ; S13, Calculate the target Q value: S14, Calculation error Used for training model parameters; S15, if If the learner achieves the target competence; If so, return to step S7 and execute sequentially until the learner reaches the target competence.

[0008] Furthermore, when selecting a knowledge point from the candidate knowledge point set as the current action, the Epsilon-Greedy algorithm is used for exploration.

[0009] Furthermore, the candidate knowledge point set obtained through the knowledge point navigation method includes: Based on the user's knowledge point learning records and knowledge structure diagram, a set of candidate knowledge points suitable for current learning is selected. The knowledge point navigation method can reduce unnecessary exploration of the model in the environment.

[0010] Given a knowledge point structure graph G, where nodes correspond to knowledge points and edges correspond to the preorder traversal between knowledge points, it is a directed acyclic graph. The set of already learned knowledge points is given. A collection of knowledge points that have not yet been learned target knowledge point set Determine the candidate knowledge point set The steps are as follows: S21, all learned knowledge points and knowledge points The edges are removed from the knowledge point structure graph G; S22, covering knowledge points not yet learned. Add to candidate knowledge point set In the middle, and delete from the knowledge point structure diagram G. and its edges; S23, Traverse all nodes in the knowledge point structure graph G, and if a set of knowledge points has not been learned, then... With the target knowledge point set If there is no path between them, then delete the unlearned knowledge point in the knowledge point structure diagram G; S24, Traverse all nodes in the knowledge point structure graph G, and store all nodes with an in-degree of 0 into the candidate knowledge point set. middle.

[0011] S25, Output the candidate knowledge point set .

[0012] Furthermore, the loss function during the training process of a deep neural network model is loss( )as follows:

[0013] in The DQN algorithm is used in the training stage. The target Q value used during the step:

[0014] in express Initial network parameters, These represent the parameters of the target network.

[0015] Furthermore, the target Q value is calculated using the following formula:

[0016] Furthermore, both initial competence and target competence are constituted by the learner's knowledge, and both initial competence and target competence are tested through exercises.

[0017] Furthermore, in deep neural network models, a state space is set up, and the states... The information contained within needs to be sufficient for the deep neural network model to determine the action with the lowest learning cost that achieves the target competence in the current state. This information includes the following four parts: learner... learners' initial competence learners' goal competence and the students' curriculum system ; After selecting the learner's information, the agent starts from the starting point and proceeds through the learner's information. learners' initial competence learners' goal competence and the students' curriculum system To obtain the current state at step t, it is represented as... ; Given the state at step t The learner selects an action from the action space to execute. In the action space, the action at step t consists of the learner and a knowledge point selected by the learner. After determining the current action, the learner selects the next course. The action at step t is represented as... , This represents the learner corresponding to step t. This represents the knowledge point selected at step t; after the agent's action at step t, the environment will calculate the reward value at step t as feedback for the current action.

[0018] Furthermore, the agent's decisions will be influenced by the reward function. The greater the improvement in a learner's competence after completing a course, the greater the reward. The reward at step t... for:

[0019] In this process, learners are assessed through two tests before studying courses containing this knowledge point. The score on the pre-learning test is... The score on the test after learning was The total score is .

[0020] Furthermore, in the low-learning-cost deep neural network model, two convolution operations are performed to extract features from all the courses the learner needs to learn. After the two convolutions, the resulting features are rearranged and used as input to a subsequent neural network containing three fully connected layers, ultimately yielding the selectable actions corresponding to the current state. The agent will select the value. The action with the largest value is taken as the action at the current step t. Compared with existing technologies, this invention has at least the following beneficial effects: By constructing a low-learning-cost deep neural network model, this invention enables each learner to progress from their initial competence to their target competence through recommended courses. The advantages of this invention are: it provides fair and personalized course recommendations for each learner; it does not rely on historical data when recommending courses, but instead uses a reinforcement learning agent to select the optimal course from a series of courses each time, maximizing the impact of the course on the learner; and because the selected course is always optimal, the learning path required for the learner to reach their target competence is minimized, significantly saving learning time. Compared with existing technologies, learners do not need to learn many courses at once, but only need to learn the fewest courses to reach their target competence. Attached Figure Description

[0021] Figure 1This is a flowchart of Embodiment 1 of the present invention; Figure 2 This is a flowchart for training deep learning networks with low learning costs; Figure 3 This is a schematic diagram of the structure of a deep learning network with low learning costs. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and examples, but the present invention is not limited to the embodiments described below.

[0023] Example 1 refer to Figure 1 The research method for intelligent knowledge point navigation in learner competency models in this embodiment includes the following steps: (1) Construction of the dataset Construct a dataset that includes multiple learners and the multiple courses each learner will take. The dataset includes the learner set, each student's course system set, initial competence, target competence, and learning cost.

[0024] The learner set L is as follows: ; Student curriculum SC as follows: ; in, It is a collection of courses for learner i.

[0025] , This indicates the total number of courses the student is required to take.

[0026] Initial competence for:

[0027] in, Indicate learner i Initial knowledge, Target competence for:

[0028] in, Indicate learner i The target knowledge.

[0029] Learning cost V The number of courses a learner needs to take to reach the target competence from the initial competence.

[0030] (2) Quantify initial competence and target competence Both initial competence and target competence are constituted by the learner's knowledge. The vector of knowledge points is denoted as: Let c represent the total number of knowledge points, which is a finite positive integer. The knowledge possessed by the learner can be tested through exercises.

[0031] (3) Construct a deep neural network model with low learning cost The described deep neural network model enables learners to improve their competencies towards a target competency after completing a series of courses, with minimal cost. Ultimately, it transforms a lengthy learning path into a shorter one, allowing learners to progress sequentially and achieve their target competency. The deep neural network model divides the learning path into two aspects: the knowledge point level and the course learning level. The knowledge point level considers which knowledge points learners need to master to reach the target competency, while the course learning level considers which courses are required to master a specific knowledge point.

[0032] In the deep neural network model, the problem to be considered is transformed into a Markov decision process, and a deep reinforcement learning agent iteratively selects the knowledge points to be chosen, thereby obtaining the desired course selection, improving students' competence and achieving the target competence. The problem can be described as consisting of states... ,action ,award A Markov Decision Process (MDP) composed of elements. This study explores how MDPs generate learning paths to achieve target competencies. During the learning process, the current state serves as the agent's input; the agent outputs an action and receives a reward for performing that action in the current state, thus stimulating the agent to learn.

[0033] state It is a set of states. This means embedding the current learner's knowledge point learning records into a vector. It indicates the first i The knowledge points that each student learns at time t. ,in It is a knowledge point embedding function. It is a record of the knowledge points learned by the learner in the course of history.

[0034] action It is a collection of actions, a single action. Corresponding to a knowledge point .

[0035] award The reward function is related to the current action and the next state, i.e. The total accumulated rewards are represented as follows:

[0036] In this model, at the knowledge point level, the initial competency information of a learner and the target knowledge points required to achieve the target competency are first obtained. Based on this information, the model recommends a currently available knowledge point to the learner. The learner then enters a course containing that knowledge point and takes quizzes before and after learning. The reward function calculates the reward the model receives for completing the current action. After learning, the recommendation strategy is optimized based on the rewards received. Subsequently, the model recommends the next knowledge point to the learner based on the rewards received, and this process is repeated continuously. Ultimately, the learner learns the target knowledge point, thereby achieving the target competency.

[0037] The general process of a low-learning-cost deep neural network model is as follows: 11) The learner already possesses a known initial competence before learning, and the learner's target competence is also known.

[0038] 12) Low-learning-cost deep neural network models recommend courses covering knowledge points that are currently suitable for learners based on their initial competence and target competence.

[0039] 13) After completing the recommended course, learners will take the course's exercises and tests. The test records will be saved, and each test will affect the learner's competence, potentially enabling them to reach a new level of competence.

[0040] 14) Calculate the learner’s progress in each course by using the learner’s initial competence and the new competence gained after completing each course, and use this as a reward.

[0041] 15) After a learner completes a course, their learning history changes. Upon receiving the new learning record and reward, the model recommends the next suitable knowledge point for the learner, who then continues learning the next course.

[0042] 16) After completing each course, add the completed course to a new course set.

[0043] 17) After the low-learning-cost deep neural network model acquires the learner's new competence, it compares the new competence with the target competence. If the target competence is achieved, the learning ends; if the target competence is not achieved, steps 13) to 16) are repeated.

[0044] 18) Learning will not end until the current learner’s new competence reaches the target competence.

[0045] (4) Knowledge Point Navigation This embodiment designs a knowledge point navigation method that, based on the user's knowledge point learning records and knowledge structure diagram, filters out a set of candidate knowledge points suitable for current learning. This knowledge point navigation method can reduce unnecessary exploration of the model within the environment.

[0046] Given a knowledge point structure graph G, where nodes correspond to knowledge points and edges correspond to the preorder traversal between knowledge points, it is a directed acyclic graph. The set of already learned knowledge points is given. A collection of knowledge points that have not yet been learned target knowledge point set Determine the candidate knowledge point set The steps are as follows: 21) All the knowledge points that have been learned and knowledge points The edges are removed from the knowledge point structure graph G; 22) Knowledge points that have not yet been learned Add to candidate knowledge point set In the middle, and delete from the knowledge point structure diagram G. and its edges; 23) Traverse all nodes in the knowledge point structure graph G. If a set of knowledge points has not been learned, then... With the target knowledge point set If there is no pathway between them, then the unlearned knowledge point is deleted from the knowledge point structure graph G.

[0047] 24) Traverse all nodes in the knowledge point structure graph G, and store all nodes with an in-degree of 0 into the candidate knowledge point set. middle.

[0048] 25) Output the candidate knowledge point set .

[0049] (5) Training deep neural network models with low learning costs The loss function for training a deep neural network model with low learning cost is loss( )as follows:

[0050] in The DQN algorithm is used in the training stage. The target Q value used during the step:

[0051] in express Initial network parameters, These represent the parameters of the target network.

[0052] The core algorithm underlying low-learning-cost deep neural network models is the DQN (Deep Q-Network) algorithm. The specific method for training low-learning-cost deep neural network models is as follows: 1) Randomly initialize the parameters F of a deep neural network model with low learning cost; 2) Initialize the experience pool H; 3) Initialize a new set of courses. This collection is used to store courses that the learner has already studied and that have improved their competence. Initially empty; 4) Randomly select a learner And acquire its target knowledge points ; 5) Initialize the knowledge point learning record and obtain the initial state. ; 6) Obtain a set of candidate knowledge points through the knowledge point navigation method.

[0053] 7) The Epsilon-Greedy algorithm is used for exploration, and a knowledge point is selected from the set of candidate knowledge points as the current action. Then proceed to the course containing that knowledge point. To learn; 8) Complete the action The newly acquired competence is compared with the initial competence to obtain the corresponding reward for the current state. Update the knowledge point learning record L and obtain the status of the next step. ; 9) Select the course at this time Put in In the set; 10) Place it into experience pool H.

[0054] 11) Status replace ; 12) Randomly select a sample from the experience replay pool H. ; 13) The target Q value is calculated as follows:

[0055] 14) Calculation error To train network F; 15) If If the learner achieves the target competence, the process ends; if Then proceed to step 7). (6) Test deep neural network models with low learning costs The test set is input into a trained, low-learning-cost deep neural network model, which outputs the course path to achieve the target competency.

[0056] (7) Performance evaluation of deep neural networks with low learning cost A low-learning-cost deep neural network model is used to assess whether each learner has achieved the target competency after completing a series of courses and to list the course path required to achieve the target competency.

[0057] Example 2 The process for training a low-learning-cost deep learning network in this embodiment includes the following steps: 1) Randomly initialize the parameters of this low-learning-cost deep neural network model; 2) Initialize the experience pool H; 3) Initialize a new set of courses. This is used to store courses that the learner has already taken and that have improved their competence. (Set) Initially empty; 4) Randomly select a learner And acquire its target knowledge points . 5) Initialize the knowledge point learning record and obtain the initial state. ; 6) Obtain a set of candidate knowledge points through knowledge point navigation.

[0058] 7) The Epsilon-Greedy algorithm is used for exploration, and a knowledge point is selected from the set of candidate knowledge points as the current action. Then proceed to the course containing that knowledge point. To learn; 8) Complete the action The newly acquired competence is compared with the initial competence to obtain the corresponding reward for the current state. Update the knowledge point learning record L and obtain the status of the next step. ; 9) Select the course at this time Put in In the set; 10) Place it into experience pool H; 11) Status replace ; 12) Randomly select a sample from the experience replay pool H. ; 13) The target Q value is calculated as follows:

[0059] 14) Calculation error To train model F; 15) If If the learner achieves the target competence, the process ends; if Then proceed to step 7). Example 3 The research method for intelligent navigation of knowledge points in the learner competency model of this embodiment includes the following steps: steps (1) to (7) are the same as in embodiment 1; The low-learning-cost deep neural network model construction in this embodiment includes the following steps: 1) The learner already possesses a known initial competence before learning. And the learner's target competence is also known.

[0060] 2) The model recommends courses covering the knowledge points that are currently suitable for learners based on their initial competence and target competence.

[0061] 3) After completing the course, learners will take the course's exercises and tests, and the test records will be saved. This test will affect the learner's competence and may enable the learner to reach a new level of competence.

[0062] 4) Calculate the learner's progress after completing each course by using the learner's initial competence and the new competence gained after completing each course, and use this as a reward for the model.

[0063] 5) After a learner completes a course, their learning history will change. Upon receiving the new learning record and reward, the model will recommend the next suitable knowledge point to the learner so they can continue learning the next course.

[0064] 6) After completing each course, add it to a new course collection.

[0065] 7) After the model acquires the learner's new competence, it compares it with the target competence. If the target competence is achieved, the learning ends; if the target competence is not achieved, steps 3 to 6 are repeated.

[0066] 8) Learning ends once the learner has achieved the target competence.

[0067] To address the challenges of improving student learning efficiency and outcomes, this paper proposes an intelligent knowledge point navigation method oriented towards learner competency models. This intelligent knowledge point navigation plays a crucial role in cultivating student competency. Firstly, by integrating and analyzing teaching data, it provides students with more precise and personalized learning paths, helping to improve their learning efficiency and effectiveness. Secondly, intelligent knowledge point navigation can provide a visual representation of the inherent connections between knowledge points, helping students build deeper understanding and further enhancing their comprehension and application abilities. Furthermore, intelligent knowledge point navigation enables teachers to better understand students' learning progress, allowing them to provide more precise guidance. Teachers can also recommend customized practice questions and reinforcement materials for incorrect answers based on students' learning status and abilities. In the era of artificial intelligence, the application of intelligent knowledge point navigation can also train students' thinking methods and enhance their digital competence.

Claims

1. A knowledge point intelligent navigation method oriented towards learner competency models, characterized in that, Based on a low-learning-cost deep neural network model, this method obtains a learner's initial competency information and the target knowledge points required to achieve the target competency at the knowledge point level. It then recommends a currently learnable knowledge point to the learner, who then enters a course containing that knowledge point. Tests are administered before and after learning, and the reward function is used to calculate the reward for completing the current action. After learning, the next knowledge point is recommended to the learner based on the reward received, and this process is repeated until the learner learns the target knowledge point and achieves the target competency. The method then provides the course path required for the learner to achieve the target competency. Low-learning-cost deep neural network models transform the entire process into a state-based process. ,action and rewards The Markov decision process is composed of various components. By exploring the Markov decision process, a curriculum path is generated to achieve the target competency learned in the course. This represents the current learner's knowledge point learning record and actions. Corresponding to a knowledge point , During the learning process, the current state serves as the input to the agent. The agent outputs an action and receives a reward for performing this action in the current state, which is used to stimulate the agent to continuously learn. The knowledge points that can be learned at present are obtained from a set of candidate knowledge points, which is obtained through a knowledge point navigation method. Both the initial competence and the target competence are composed of the learner's knowledge, which is acquired through exercise tests.

2. The intelligent knowledge point navigation method for learner competency models according to claim 1, characterized in that, The specific methods for training deep neural network models with low learning costs include the following steps: S1, randomly initializes the parameters of a deep neural network model with low learning cost; S2, Initialize the experience pool H; S3, Initialize the new course set This collection is used to store courses that a learner has already studied and that have improved their competence. Initially empty; S4, randomly select a learner And acquire its target knowledge points ; S5, Initialize knowledge point learning records and obtain the initial state. ; S6, obtain a set of candidate knowledge points through the knowledge point navigation method; S7: Select a knowledge point from the candidate knowledge point set as the current action. Then proceed to the course containing that knowledge point. To learn; S8 will complete the action. The newly acquired competence is compared with the initial competence to obtain the corresponding reward for the current state. Update the knowledge point learning record L and obtain the status of the next step. ; S9, select the course at this time Put in In the set; S10, will Place it into experience pool H; S11, using state replace ; S12, Randomly collect a sample from the experience replay pool H. ; S13, Calculate the target Q value: S14, Calculation error Used for training model parameters; S15, if If the learner achieves the target competence, then the learner will achieve the target competence; if Then return to step S7 and execute sequentially until the learner reaches the target competency. As initial competence, , Indicate learner i Initial knowledge, For target competence, , Indicate learner i The target knowledge.

3. The intelligent knowledge point navigation method for learner competency models according to claim 2, characterized in that, When selecting a knowledge point from the candidate knowledge point set as the current action, the Epsilon-Greedy algorithm is used for exploration.

4. The intelligent knowledge point navigation method for learner competency models according to claim 2, characterized in that, The candidate knowledge point set obtained through the knowledge point navigation method includes: Based on the user's knowledge point learning records and knowledge structure diagram, a set of candidate knowledge points suitable for learning is selected. The knowledge point navigation method can reduce unnecessary exploration of the model in the environment. Given a knowledge point structure graph G, where nodes correspond to knowledge points and edges correspond to the preorder traversal between knowledge points, it is a directed acyclic graph. The set of already learned knowledge points is also given. A collection of knowledge points that have not yet been learned target knowledge point set Determine the set of candidate knowledge points The steps are as follows: S21, all learned knowledge points and knowledge points The edges are removed from the knowledge point structure graph G; S22, covering knowledge points not yet learned. Add to candidate knowledge point set In the middle, and delete from the knowledge point structure diagram G. and its edges; S23, Traverse all nodes in the knowledge point structure graph G, and if a set of knowledge points has not been learned, then... With the target knowledge point set If there is no path between them, then delete the unlearned knowledge point in the knowledge point structure diagram G; S24, Traverse all nodes in the knowledge point structure graph G, and store all nodes with an in-degree of 0 into the candidate knowledge point set. middle; S25, Output the candidate knowledge point set .

5. The intelligent knowledge point navigation method for learner competency models according to claim 2, characterized in that, Low-learning-cost deep neural network models are based on the DQN algorithm. The loss function during the training process of deep neural network models is loss( )as follows: in It is a deep neural network model with low learning cost in the training stage. The target Q value used during the step: in express Initial network parameters, These represent the parameters of the target network.

6. The intelligent knowledge point navigation method for learner competency models according to claim 5, characterized in that, The target Q value is calculated using the following formula: 。 7. The intelligent knowledge point navigation method for learner competency models according to claim 1, characterized in that, Both initial competence and target competence are constituted by the learner's knowledge, and both initial competence and target competence are tested through exercises.

8. The intelligent knowledge point navigation method for learner competency models according to claim 1, characterized in that, In a deep neural network model, a state space is set up, and the states are... The information contained within needs to be sufficient for the deep neural network model to determine the action with the lowest learning cost that achieves the target competence in the current state. This information includes the following four parts: learner... learners' initial competence learners' goal competence and the students' curriculum system ; After selecting the learner's information, the agent starts from the starting point and proceeds through the learner's information. learners' initial competence learners' goal competence and the students' curriculum system To obtain the current state at step t, it is represented as... ; Given the state at step t The learner selects an action from the action space to execute. In the action space, the action at step t consists of the learner and a knowledge point selected by the learner. After determining the current action, the learner selects the next course. The action at step t is represented as... , This represents the learner corresponding to step t. This represents the knowledge point selected at step t; after the agent's action at step t, the environment will calculate the reward value at step t as feedback for the current action.

9. The intelligent knowledge point navigation method for learner competency models according to claim 1, characterized in that, The agent's decisions will be influenced by the reward function. The greater the improvement in a learner's competence after completing a course, the greater the reward. The reward at step t... for: Before learning a course containing this knowledge point, learners will be assessed through two tests. The score on the pre-learning test will be... The score on the test after learning was The total score is .

10. The intelligent knowledge point navigation method for learner competency models according to claim 1, characterized in that, In a low-learning-cost deep neural network model, two convolutional operations are performed to extract features from all the lessons the learner is to learn. After the two convolutional layers, the resulting features are rearranged and used as input to a subsequent neural network containing three fully connected layers, ultimately yielding the selectable actions for the current state. The agent will select the value. The action with the largest value is taken as the action at the current step t. .

Images