Learning path recommendation method and device, electronic equipment and storage medium
By constructing project hierarchical maps and the collaborative method of agents, the learning paths are dynamically adjusted, and the problem of difficulty is not considered in the existing technology is solved, achieving a smoother and more efficient learning path recommendation.
Patent Information
- Application Number
- CN202411164230.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-22
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-08-22
AI Technical Summary
The existing learning path recommendation method fails to fully consider the difficulty of the learning project, resulting in the recommended learning path not being smooth enough, resulting in low efficiency and bumpy learning paths.
By constructing project hierarchical maps, using neural networks to aggregate high-order information and difficulty information between learning projects, combining high-level agents and low-level agents, dynamically adjusting the learning path, and gradually selecting appropriate difficulty levels based on the user's current knowledge status and number of exercises practice items until mastering the target knowledge items.
A smoother learning path is achieved, learning efficiency is improved, learners can gradually master knowledge under appropriate difficulty, and avoid learning disabilities that waste time and excessively difficult on the content they have mastered.
Smart Images

Figure CN119066240A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data mining technology, and in particular to a learning path recommendation method and device, electronic equipment, and storage medium. Background Art
[0002] With the development of educational technology and the popularity of online learning platforms, learners have more convenient access to learning resources. However, with the explosive growth of knowledge and learning resources in the information age, it also brings new pain points for people to acquire new knowledge, that is, how learners can accurately obtain learning resources that suit their needs from a large number of complex learning resources and learn efficiently in an appropriate learning order.
[0003] Therefore, the current technology for recommending learning paths for learners has emerged. The current method for learning paths is mainly to target the various knowledge items that learners need to learn, that is, the knowledge points or skills to be learned, etc. According to the predecessor and successor relationship of each knowledge item, that is, the learning order relationship of the knowledge items, the knowledge items that learners currently need to learn are determined in sequence, and all the practice items that have an examination relationship with the knowledge items are provided to learners for practice, that is, the associated items are provided to learners for practice, so that they can learn and master the corresponding knowledge items.
[0004] However, since this method only takes into account the predecessor and successor relationships and the inspection relationship between projects, and does not take into account the difficulty of the projects, the recommended learning path is not smooth enough, and the learning path is "inefficient", that is, learning projects of different difficulty levels with the same learning amount, resulting in spending extra time on knowledge that has already been mastered or failing to effectively master the knowledge after completing the learning amount, and a "bumpy" learning path, that is, after mastering more difficult exercises, you will still be recommended to learn projects with lower difficulty. Summary of the invention
[0005] Based on the above-mentioned deficiencies of the prior art, the present application provides a learning path recommendation method and device, an electronic device, and a storage medium to solve the problem that the learning path recommended by the prior art is not smooth enough.
[0006] In order to achieve the above objectives, this application provides the following technical solutions:
[0007] The first aspect of the present application provides a learning path recommendation method, comprising:
[0008] According to the association relationship between various learning items, a project hierarchical diagram is constructed; wherein the learning items include multiple knowledge items and their corresponding exercise items; the association relationship includes a predecessor-successor relationship and an inspection relationship;
[0009] Aggregating high-level information and difficulty information between the learning items through a hierarchical graph neural network to obtain item representations of each of the exercise items;
[0010] The high-level agent determines the current knowledge item from each target knowledge item on the item hierarchy diagram according to the user's current knowledge state and the number of exercises in the previous stage, and determines the current initial difficulty and the current limit number of exercises, and transmits them to the low-level agent;
[0011] The low-level agent gradually selects current practice items of corresponding difficulty from the practice items corresponding to the current knowledge items based on the current knowledge status of the user, starting from the current initial difficulty, and provides them to the user for learning, and accumulates the number of exercises in the current stage and updates the current knowledge status of the user based on the item representation and learning feedback of the current practice items, until the number of exercises in the current stage reaches the current exercise limit number, or the current knowledge status of the user reaches the status threshold;
[0012] The low-level agent feeds back the most recently updated number of exercises in the current stage as the number of exercises in the previous stage and the most recently updated current knowledge status of the user to the high-level agent, so as to return and execute the high-level agent to determine the current knowledge item from the various target knowledge items on the project hierarchy diagram according to the user's current knowledge status and the number of exercises in the previous stage, as well as the current initial difficulty and the current limit on the number of exercises until the user masters each of the target knowledge items.
[0013] Optionally, in the above-mentioned learning path recommendation method, the aggregating high-level information and difficulty information between the learning items through a hierarchical graph neural network to obtain the item representation of each of the practice items includes:
[0014] Based on the historical learning records, the ratio of the number of incorrect answers to the total number of answers for each learning item is multiplied by the corresponding preset difficulty level to obtain the difficulty of each learning item;
[0015] Generate a knowledge item embedding matrix and an exercise item embedding matrix, and generate a knowledge item difficulty embedding matrix and an exercise item difficulty embedding matrix according to the difficulty of each of the learning items; wherein the knowledge item embedding matrix and the exercise item embedding matrix are respectively used to represent the full amount of the knowledge items and the exercise items;
[0016] For each of the knowledge items, respectively, the item vector corresponding to the knowledge item in the knowledge item embedding matrix and the difficulty vector corresponding to the knowledge item in the knowledge item difficulty embedding matrix are concatenated, multiplied by a first weight matrix, and added with a first bias term to obtain a combined embedding vector of each of the knowledge items; and for each of the practice items, the item vector corresponding to the practice item in the practice item embedding matrix and the difficulty vector corresponding to the practice item in the practice item difficulty embedding matrix are concatenated, multiplied by a second weight matrix, and added with a second bias term to obtain a combined embedding vector of each of the practice items;
[0017] For each of the exercise items, the combined embedding vector of the exercise item is connected with the mean of the combined embedding vectors of its adjacent knowledge items to obtain the item representation of each of the exercise items.
[0018] Optionally, in the above-mentioned learning path recommendation method, the high-level agent determines the current knowledge item from each target knowledge item on the item hierarchy diagram according to the user's current knowledge state and the number of exercises in the previous stage, and determines the current initial difficulty and the current limit number of exercises, including:
[0019] The high-level agent connects the current knowledge state of the user with the encoding vectors of each target knowledge item to obtain the current state of the high-level agent;
[0020] Determine the current knowledge item from the current knowledge item dynamic space based on the current state of the high-level agent using a proximal strategy optimization algorithm; wherein the current knowledge item dynamic space is composed of the target knowledge item on the item hierarchical graph;
[0021] The current initial difficulty is calculated using the user's current knowledge status and a custom correct answer probability;
[0022] The current practice limit number of times is calculated by a trained multi-layer perceptron based on the user's current knowledge status, the practice number of the previous stage, the practice limit number of the previous stage, and the difficulty of the current knowledge item.
[0023] Optionally, in the above-mentioned learning path recommendation method, starting from the current initial difficulty, the low-level agent gradually selects current practice items of corresponding difficulty from the practice items corresponding to the current knowledge items according to the current knowledge state of the user, provides them to the user for learning, accumulates the number of exercises in the current stage, and updates the current knowledge state of the user according to the item representation and learning feedback of the current practice items, until the number of exercises in the current stage reaches the current exercise limit number, or the current knowledge state of the user reaches the state threshold, including:
[0024] Selecting, by the low-level agent, from the various exercise items corresponding to the current knowledge item, an exercise item with a difficulty of the current initial difficulty as the current exercise item, and providing it to the user for learning;
[0025] Accumulating the number of exercises in the current stage and updating the current knowledge state of the user according to the project representation and learning feedback of the current exercise project;
[0026] Determine whether the accumulated number of exercises in the current stage reaches the current exercise limit number, or whether the current knowledge status of the user reaches a status threshold;
[0027] If it is determined that the accumulated number of exercises in the current stage does not reach the current exercise limit number, and the current knowledge state of the user does not reach the state threshold, the current knowledge state of the user is connected with the encoding vector of the current knowledge item to obtain the current state of the low-level agent;
[0028] Using the actor-critic algorithm, based on the current state of the low-level agent, a new current learning item is selected from the various practice items corresponding to the current knowledge item, provided to the user for learning, and the current knowledge state of the user is updated according to the item representation and learning feedback of the current practice item and the accumulated number of practice times in the current stage;
[0029] If it is determined that the accumulated number of exercises in the current stage reaches the current exercise limit, or the current knowledge status of the user reaches the status threshold, the low-level agent will use the currently updated number of exercises in the current stage as the number of exercises in the previous stage, and feed back the currently updated number of exercises in the current stage and the current knowledge status of the user to the high-level agent.
[0030] A second aspect of the present application provides a learning path recommendation device, comprising:
[0031] A construction module is used to construct a project hierarchical diagram according to the association relationship between various learning projects; wherein the learning projects include multiple knowledge projects and their corresponding exercise projects; the association relationship includes a predecessor-successor relationship and an inspection relationship;
[0032] An aggregation module, used to aggregate high-level information and difficulty information between the learning items through a hierarchical graph neural network to obtain item representations of each of the exercise items;
[0033] A high-level agent module, used to determine the current knowledge item from each target knowledge item on the item hierarchy diagram according to the user's current knowledge state and the number of exercises in the previous stage through the high-level agent, as well as the current initial difficulty and the current limit number of exercises, and pass them to the low-level agent;
[0034] A low-level agent module is used to select, from the current initial difficulty, through the low-level agent, a current exercise item of corresponding difficulty from each exercise item corresponding to the current knowledge item according to the current knowledge state of the user, and provide it to the user for learning, and accumulate the number of exercises in the current stage and update the current knowledge state of the user according to the item representation and learning feedback of the current exercise item, until the number of exercises in the current stage reaches the current exercise limit number, or the current knowledge state of the user reaches a state threshold;
[0035] The low-level agent feeds back the most recently updated number of exercises in the current stage as the number of exercises in the previous stage and the most recently updated current knowledge status of the user to the high-level agent, so as to return and execute the high-level agent to determine the current knowledge item from the various target knowledge items on the project hierarchy diagram according to the user's current knowledge status and the number of exercises in the previous stage, as well as the current initial difficulty and the current limit on the number of exercises until the user masters each of the target knowledge items.
[0036] Optionally, in the above-mentioned learning path recommendation device, the aggregation module includes:
[0037] A difficulty calculation module, for obtaining the difficulty of each learning item by multiplying the ratio of the number of incorrect answers to the total number of answers for each learning item by the corresponding preset difficulty level based on historical learning records;
[0038] A matrix generation module, used to generate a knowledge item embedding matrix and an exercise item embedding matrix, and to generate a knowledge item difficulty embedding matrix and an exercise item difficulty embedding matrix according to the difficulty of each of the learning items; wherein the knowledge item embedding matrix and the exercise item embedding matrix are respectively used to represent the full amount of the knowledge items and the exercise items;
[0039] A fusion module, for respectively, for each of the knowledge items, concatenating the item vector corresponding to the knowledge item in the knowledge item embedding matrix with the difficulty vector corresponding to the knowledge item in the knowledge item difficulty embedding matrix, multiplying the result of the concatenation operation by a first weight matrix, and adding the result with a first bias term, to obtain a combined embedding vector of each of the knowledge items; and for respectively, for each of the practice items, concatenating the item vector corresponding to the practice item in the practice item embedding matrix with the difficulty vector corresponding to the practice item in the practice item difficulty embedding matrix, multiplying the result of the concatenation operation by a second weight matrix, and adding the result with a second bias term, to obtain a combined embedding vector of each of the practice items;
[0040] The combining module is used to connect the combined embedding vector of each exercise item with the mean of the combined embedding vectors of its adjacent knowledge items for each exercise item, so as to obtain the item representation of each exercise item.
[0041] Optionally, in the above-mentioned learning path recommendation device, when the high-level agent module executes the step of determining the current knowledge item from each target knowledge item on the item hierarchy diagram according to the user's current knowledge state and the number of exercises in the previous stage through the high-level agent, and determining the current initial difficulty and the current limit number of exercises, it is used to:
[0042] The high-level agent connects the current knowledge state of the user with the encoding vectors of each target knowledge item to obtain the current state of the high-level agent;
[0043] Determine the current knowledge item from the current knowledge item dynamic space based on the current state of the high-level agent using a proximal strategy optimization algorithm; wherein the current knowledge item dynamic space is composed of the target knowledge item on the item hierarchical graph;
[0044] The current initial difficulty is calculated using the user's current knowledge status and a custom correct answer probability;
[0045] The current practice limit number of times is calculated by a trained multi-layer perceptron based on the user's current knowledge status, the practice number of the previous stage, the practice limit number of the previous stage, and the difficulty of the current knowledge item.
[0046] Optionally, in the above-mentioned learning path recommendation device, the low-level agent module executes the step of selecting, from the current initial difficulty, the current practice items of corresponding difficulty from the practice items corresponding to the current knowledge items according to the current knowledge state of the user through the low-level agent, and providing them to the user for learning, and accumulating the number of exercises in the current stage and updating the current knowledge state of the user according to the item representation and learning feedback of the current practice items, until the number of exercises in the current stage reaches the current exercise limit number, or the current knowledge state of the user reaches the state threshold, and is used to:
[0047] Selecting, by the low-level agent, from the various exercise items corresponding to the current knowledge item, an exercise item with a difficulty of the current initial difficulty as the current exercise item, and providing it to the user for learning;
[0048] Accumulating the number of exercises in the current stage and updating the current knowledge state of the user according to the project representation and learning feedback of the current exercise project;
[0049] Determine whether the accumulated number of exercises in the current stage reaches the current exercise limit number, or whether the current knowledge status of the user reaches a status threshold;
[0050] If it is determined that the accumulated number of exercises in the current stage does not reach the current exercise limit number, and the current knowledge state of the user does not reach the state threshold, the current knowledge state of the user is connected with the encoding vector of the current knowledge item to obtain the current state of the low-level agent;
[0051] Using the actor-critic algorithm, based on the current state of the low-level agent, a new current learning item is selected from the various practice items corresponding to the current knowledge item, provided to the user for learning, and the current knowledge state of the user is updated according to the item representation and learning feedback of the current practice item and the accumulated number of practice times in the current stage;
[0052] If it is determined that the accumulated number of exercises in the current stage reaches the current exercise limit, or the current knowledge status of the user reaches the status threshold, the low-level agent will use the currently updated number of exercises in the current stage as the number of exercises in the previous stage, and feed back the currently updated number of exercises in the current stage and the current knowledge status of the user to the high-level agent.
[0053] A third aspect of the present application provides an electronic device, including:
[0054] Memory and processor;
[0055] Wherein, the memory is used to store programs;
[0056] The processor is used to execute the program, and when the program is executed, it is specifically used to implement the learning path recommendation method as described in any one of the above.
[0057] A fourth aspect of the present application provides a computer storage medium for storing a computer program, wherein when the computer program is executed, it is used to implement the learning path recommendation method as described in any one of the above.
[0058] The present application provides a learning path recommendation method, which constructs a project hierarchical graph according to the association relationship between each learning project. Among them, the learning project includes multiple knowledge projects and their corresponding various exercise projects. The association relationship includes a predecessor-successor relationship and an inspection relationship. Then, the high-order information and difficulty information between the learning projects are aggregated through a hierarchical graph neural network to obtain the project representation of each exercise project, so as to represent the exercise project through the representation that integrates the difficulty information, so as to take into account the difficulty of the project when recommending the path. Then, the high-level intelligent agent determines the current knowledge project from the various target knowledge projects on the project hierarchical graph according to the user's current knowledge state and the number of exercises in the previous stage, and determines the current initial difficulty and the current exercise limit number, and passes it to the low-level intelligent agent. Then, the low-level intelligent agent, starting from the current initial difficulty, gradually selects the current exercise project of the corresponding difficulty from the various exercise projects corresponding to the current knowledge project according to the user's current knowledge state, and provides it to the user for learning, and accumulates the number of exercises in the current stage and the project representation and learning feedback of the current exercise project, and updates the user's current knowledge state until the number of exercises in the current stage reaches the current exercise limit number, or the user's current knowledge state reaches the state threshold. The low-level agent uses the most recently updated number of exercises in the current stage as the number of exercises in the previous stage, and feeds back the most recently updated current knowledge state of the user to the high-level agent, so as to return to execute the high-level agent for the next stage of learning. Thus, through the high-level agent and the low-level agent, and the interaction between the two, a difficulty-driven method of recommending knowledge items to learn and exercise items under the knowledge items is implemented, thereby ensuring that the recommended learning path is smoother and improving learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.
[0060] Figure 1A flowchart of a learning path recommendation method provided in an embodiment of the present application;
[0061] Figure 2 A flowchart of a method for calculating an item representation of an exercise item provided in an embodiment of the present application;
[0062] Figure 3 A flowchart of a method for determining a current knowledge item, a current initial difficulty, and a current practice limit number provided in an embodiment of the present application;
[0063] Figure 4 A flowchart of a method for recommending exercise items to a low-level agent provided in an embodiment of the present application;
[0064] Figure 5 A schematic diagram of a system architecture for implementing a learning path recommendation method provided in an embodiment of the present application;
[0065] Figure 6 A schematic diagram of the architecture of a learning path recommendation device provided in an embodiment of the present application;
[0066] Figure 7 A schematic diagram of the architecture of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0067] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0068] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.
[0069] The present application embodiment provides a learning path recommendation method, such as Figure 1 As shown, the following steps are included:
[0070] S101. Construct a project hierarchy diagram based on the relationship between various learning projects.
[0071] Among them, learning projects include multiple knowledge projects and their corresponding practice projects. The association relationship includes predecessor-successor relationship and examination relationship. Knowledge projects generally refer to knowledge points or skills, such as addition, multiplication, calculus, etc. There is a predecessor-successor relationship between knowledge projects, such as addition is the predecessor of multiplication, and multiplication is the successor of addition, which means that addition needs to be learned before learning multiplication. Practice projects refer to the practice content set up to master the corresponding knowledge projects, which are practical applications or problems with an examination relationship with the knowledge projects, such as the problem "2+3" examines addition. Through practice projects, learners can learn and master the corresponding knowledge projects. At the same time, learners can get instant feedback every time they interact with practice projects. The feedback is usually right or wrong or a score. Learners can understand their learning situation through feedback.
[0072] Therefore, in the embodiment of the present application, in order to fully consider the learning projects, a project hierarchical graph is first constructed based on the association relationship between the learning projects. The project hierarchical graph can be specifically expressed as HG = {V, E}. Among them, V represents the set of projects, which is all the nodes in the hierarchical graph, including knowledge projects and practice projects; E represents the set of project relationships, which is all the edges in the hierarchical graph, including the predecessor and successor relationships between knowledge projects and the examination relationship between knowledge projects and practice projects.
[0073] Specifically, the edge (c i , c j ) represents knowledge item c i It is a knowledge project c j Precursor knowledge and learning foundation; the edge connecting the exercise items and knowledge items in E (e m , c j ) indicates exercise item e m The knowledge item c j .
[0074] S102. Aggregate high-level information and difficulty information between learning items through a hierarchical graph neural network to obtain item representations of various exercise items.
[0075] It should be noted that in order to better consider the relationship between learning items and the difficulty of learning items in the process of path determination, in an embodiment of the present application, high-order information and difficulty information between learning items are aggregated through a hierarchical graph neural network to obtain project representations of various practice items, and the project representations of the practice items are used to represent the nodes of each practice item in the project hierarchical graph, so that the relationship between learning items and the difficulty can be taken into account when exploring the learning path.
[0076] Specifically, each learning item and the difficulty of each learning item are represented by a corresponding representation vector, and the two are fused to obtain a vector with integrated difficulty characteristics. Then, the vector representation of the integrated difficulty characteristics of the practice item and the associated knowledge item are fused to obtain the item representation of each practice item.
[0077] Optionally, in another embodiment of the present application, a specific implementation of step S102 is as follows: Figure 2 As shown, including:
[0078] S201. Based on historical learning records, multiply the ratio of the number of incorrect answers to the total number of answers for each learning item by the corresponding preset difficulty level to obtain the difficulty of each learning item.
[0079] Specifically, historical learning records of a large number of users for various learning projects are collected, so that the difficulty of each learning project can be calculated based on the accuracy of the answers to each historical learning project. Among them, the total number of people who answered an exercise project is the total number of people who have interacted with the exercise project, that is, the total number of people who answered more than one exercise project. Correspondingly, the number of incorrect answers to an exercise project is the number of people who did not answer the exercise project correctly. The total number of answers to knowledge projects is the total number of people who answered various exercise projects associated with the knowledge project. The number of incorrect answers to knowledge projects is the total number of people who did not answer various exercise projects associated with the knowledge project correctly. Therefore, the calculation method of the difficulty of a learning project can be specifically as follows:
[0080]
[0081] Where M represents the total number of people who answered the questions in the learning project; a m =0 means that the learner cannot answer the learning item correctly, that is, the learner who answers correctly; p Indicates preset difficulty levels.
[0082] S202, generating a knowledge item embedding matrix and an exercise item embedding matrix, and generating a knowledge item difficulty embedding matrix and an exercise item difficulty embedding matrix according to the difficulty of each learning item.
[0083] Among them, the knowledge item embedding matrix and the practice item embedding matrix are used to represent the full amount of knowledge items and practice items respectively.
[0084] S203. For each knowledge item, respectively, concatenate the item vector corresponding to the knowledge item in the knowledge item embedding matrix with the difficulty vector corresponding to the knowledge item in the knowledge item difficulty embedding matrix, multiply the result by the first weight matrix, and add the result with the first bias term to obtain the combined embedding vector of each knowledge item; and for each practice item, concatenate the item vector corresponding to the practice item in the practice item embedding matrix with the difficulty vector corresponding to the practice item in the practice item difficulty embedding matrix, multiply the result by the second weight matrix, and add the result with the second bias term to obtain the combined embedding vector of each practice item.
[0085] Therefore, the calculation methods of the combined embedding vector of the knowledge items and the combined embedding vector of the exercise items are:
[0086]
[0087] in, and denote the first weight matrix and the second weight matrix respectively; and Respectively represent the item vector and the corresponding difficulty vector corresponding to the i-th knowledge item; and They represent the item vector and the difficulty vector corresponding to the i-th exercise item respectively; b L and b P denote the first bias term and the second bias term respectively; Represents a concatenation operation.
[0088] S204 , for each exercise item, respectively, concatenate the combined embedding vector of the exercise item with the mean of the combined embedding vectors of its adjacent knowledge items to obtain the item representation of each exercise item.
[0089] Specifically, in the embodiment of the present application, simplified average aggregation is used to fuse the features of the knowledge items associated with the practice items, which can be specifically expressed as:
[0090]
[0091] in, Indicates exercise item e j The aggregated feature embedding, N j Indicates e j Neighbor nodes of knowledge items on the hierarchical graph HG.
[0092] S103. The high-level intelligent agent determines the current knowledge item from each target knowledge item on the project hierarchy diagram according to the user's current knowledge status and the number of exercises in the previous stage, and determines the current initial difficulty and the current limit number of exercises.
[0093] It should be noted that in order to recommend an efficient and smooth learning path for learners while considering the difficulty of the project, the present invention designs a difficulty-driven hierarchical reinforcement learning framework based on the hierarchical graph constructed in the text. Among them, the framework includes two agents: a high-level agent is responsible for selecting the knowledge to be learned in the next stage at the knowledge project layer. A low-level agent is responsible for selecting the next project to be practiced in the current stage at the practice project layer. At the same time, in order to ensure effective coordination between the two agents and realize effective learning path construction, a communication mechanism between the two agents is implemented. Therefore, the high-level agent determines the initial difficulty of the practice project, and the low-level agent controls the maximum number of practice times of the learner based on the learning project difficulty information transmitted from the high-level agent.
[0094] Current learning path recommendations all use deep knowledge tracking methods to evaluate learners' knowledge status by only considering the association of items. However, the current approach ignores the comprehensive use of item difficulty. Therefore, in this embodiment of the application, a knowledge tracking method DIMKT for learning difficulty matching is introduced to consider the impact of item difficulty on learners' cognitive changes. In addition, the item representation of the exercise items obtained in the previous section is used to replace the x in the original method. j To represent the practice items that interact with learners.
[0095] Therefore, through the high-level intelligent agent, based on the user's current knowledge status, that is, the status of the user when learning a knowledge item, it can be determined whether to learn the next knowledge item, or to return to the previous knowledge item for learning, or to continue learning the current knowledge item.
[0096] In order to allow users to start learning from practice items of difficulty suitable for their current learning situation and make learning more efficient, in an embodiment of the present application, the difficulty of starting learning in the current stage, that is, the current initial difficulty, is determined based on the number of exercises in the previous stage.
[0097] Furthermore, considering that learners cannot be allowed to practice the same knowledge item indefinitely to master the knowledge item, this approach is counteracting and unreasonable. When they practice a knowledge item many times and still cannot master it, it means that they have not effectively mastered the knowledge item in the previous order and need to learn again, rather than practicing the same knowledge item indefinitely. Therefore, in the embodiment of the present application, a limit on the number of exercises is set. Moreover, each learner has a different learning level and speed, so the limit on the number of exercises is personalized and dynamically adjusted. Therefore, in the embodiment of the present application, the current limit on the number of exercises is determined based on the current knowledge state and the number of exercises in the previous stage.
[0098] Optionally, in another embodiment of the present application, a specific implementation of step S103 is as follows: Figure 3 As shown, the following steps are included:
[0099] S301. Connect the user's current knowledge state with the encoding vectors of each target knowledge item through a high-level intelligent agent to obtain the current state of the high-level intelligent agent.
[0100] Specifically, the state of the high-level intelligent agent includes the learner's target knowledge, that is, the various knowledge items that the learner needs to learn, and the learner's state at the end of the previous stage. Since the learning of the next node has not yet started at this time, the knowledge state at the end of the previous stage is the current knowledge state.
[0101] Optionally, the user's target knowledge can be encoded by multi-hot encoding as G = {0, 1} K . Where k is the number of knowledge items; the position corresponding to the user's target knowledge is 1, otherwise it is set to 0. Therefore, the calculation process of the current state of the high-level intelligent agent can be specifically expressed as:
[0102]
[0103] Among them, h i-1 It is the knowledge status at the end of the previous stage.
[0104] S302. Determine the current knowledge item from the current knowledge item dynamic space based on the current state of the high-level intelligent agent using the proximal strategy optimization algorithm.
[0105] The current knowledge item dynamic space is composed of target knowledge items on the item hierarchical diagram, which is an adaptive and dynamic knowledge action space that can dynamically adjust the knowledge items to be determined according to the learning situation.
[0106] Specifically, the action of the high-level intelligent agent refers to recommending appropriate knowledge items for the current learning stage. Specifically, the A star algorithm can be used to generate the shortest path between the starting knowledge item and the target knowledge item as a reference path. When generating the learning path later, it is adjusted according to the learner's real-time learning progress. If the learning goes smoothly, continue along the reference path. Otherwise, search with the current knowledge item as the root node, and then learn step by step from near to far according to the distance from the current item. Whether the learning is smooth is judged by evaluating whether the learner has reached the expected mastery state threshold after completing a certain number of practice items, that is, whether h is satisfied. i-1 ≥threshold. Therefore, the candidate action space D can be dynamically determined L .
[0107] After determining the current dynamic space, knowledge items can be selected in the dynamic space according to the selection strategy. In the embodiment of the present application, a proximal strategy optimization algorithm is used to select knowledge items in the action space D L Select the most appropriate knowledge item as the current knowledge item.
[0108] Specifically, a policy network is used as an actor to extract Output action probabilities and use a value network As a critic, it evaluates the expected return of each state. L and are the parameters of the relevant network. Specifically, the probability distribution of knowledge items and each state The expected return under is calculated as follows:
[0109]
[0110] Among them, FC represents the fully connected layer.
[0111] During training, the mean square error loss is used to train the critic’s value network. The loss function is as follows:
[0112]
[0113] Among them, γ i Is a discount factor that discounts the importance of future rewards when calculating cumulative rewards, making the reward weight matrix farther away from the current time lower. A higher discount factor (close to 1) means more emphasis on long-term rewards, while a lower discount factor (close to 0) pays more attention to recent rewards. Represents the reward value of agent L at time t+i.
[0114] At the same time, for the training of the actor strategy network, the advantage value of each step is calculated first. The specific calculation method is as follows:
[0115]
[0116] Then, during the actor training process, the data of one round is divided into multiple iterations for training. Specifically, the actor interacting at the kth iteration is represented as Next, we use the proximal strategy to optimize the truncation loss for training. The loss function is as follows:
[0117]
[0118] Among them, clip(x,l,h)=max(min(x,h),l), which means that x is limited to [l,h]. This can make the training more stable and efficient.
[0119] In the future, the learning benefits of learners can be maximized, thereby achieving the best learning effect. Therefore, in the embodiment of this application, the reward function used for the high-level intelligent agent is:
[0120]
[0121] Among them, i is the last learning stage, and the value is E p ; if i is not the last learning stage, the value is 0. p The calculation is as follows:
[0122]
[0123] Among them, E s is the test score of the learner on the target knowledge before learning, E e is the test score of the learner on the target knowledge after completing the learning path, E sup A full score on the target knowledge test.
[0124] S303: Calculate the current initial difficulty by using the user's current knowledge status and the customized correct answer probability.
[0125] Since the current knowledge status reflects the current knowledge level of the user, it can be used to determine what level of difficulty the user should start learning next, which is most suitable for the user's current learning situation, avoiding wasting unnecessary time on simpler content that has already been mastered, and avoiding excessive difficulty that prevents new knowledge from being learned. The customized correct answer probability can reflect the probability of correctly answering the practice items, so it also has a high reference for the difficulty of selecting practice items. Therefore, the current initial difficulty is calculated using the user's current knowledge status and the customized correct answer probability. Specifically, the current initial difficulty calculation process can be as follows:
[0126]
[0127] Among them, h i represents the learner’s knowledge state at the current learning stage i, P i Represents the custom probability that a learner answers an exercise item correctly.
[0128] S304, using the trained multi-layer perceptron, based on the user's current knowledge status, the number of exercises in the previous stage, the current exercise limit number in the previous stage, and the difficulty of the current knowledge item, calculate the current exercise limit number.
[0129] It should be noted that, since the exercise limit number of the previous stage reflects the user's learning state in the previous stage, and the learning state usually does not change significantly, it will affect the number of times that the current exercise needs to be practiced. Therefore, in the embodiment of the present application, the current exercise limit number of the previous stage will also be considered. The difficulty of the current knowledge project will obviously also affect the number of times that the normal exercise needs to be practiced. The higher the difficulty, the more times the exercise needs to be practiced. Among them, the difficulty of the current knowledge project can be calculated in the manner of step S201, which will not be repeated here.
[0130] Therefore, in the embodiment of the present application, the calculation method for the current exercise limit number is as follows:
[0131] τ i =f(p i-1 ,h i-1 ,τ i-1 ,DL i )
[0132] Among them, p i-1 and h i-1 Respectively represent the number of exercises the student has done in the previous learning stage and the knowledge status at the end of the previous stage. i-1 and DL i They represent the number of exercises in the previous stage and the difficulty of the current knowledge item. f is a multi-layer perceptron, which is a trainable network. i-1 ,h i-1 , τ i-1 , DL i The historical data is used for training.
[0133] S104, determining whether the user has mastered various target knowledge items.
[0134] Optionally, it may be determined whether the user has learned and mastered all target knowledge items based on the user's current knowledge status and current knowledge items.
[0135] If it is determined that the user has mastered each item of target knowledge, then step S105 is executed. If it is determined that the user has not mastered each item of target knowledge, then step S106 is executed.
[0136] S105. End the path recommendation for each item of target knowledge.
[0137] S106. Pass the current knowledge item, the current initial difficulty, and the current practice limit to the lower-level agent.
[0138] S107. Starting from the current initial difficulty, the low-level intelligent agent gradually selects current practice items of corresponding difficulty from the practice items corresponding to the current knowledge items according to the user's current knowledge status, and provides them to the user for learning. The number of exercises in the current stage is accumulated, and the current knowledge status of the user is updated according to the item representation of the current practice items and learning feedback, until the number of exercises in the current stage reaches the current exercise limit, or the current knowledge status of the user reaches the status threshold.
[0139] Specifically, after the low-level agent has finished transmitting the current knowledge item, the current initial difficulty, and the current practice limit to the high-level agent, it will use the current initial difficulty as the initial difficulty to select the current practice item of the corresponding difficulty from the various practice items corresponding to the current knowledge item, and provide it to the user for learning. In addition, to avoid the number of exercises exceeding the limit, the number of exercises in the current stage will be accumulated, that is, from the time the information is received, the number of exercises in the current stage will start from zero, and each time an exercise item is provided to the user, the number of exercises in the current stage will be accumulated by 1. In order to continuously adjust the difficulty of the next exercise item to be selected according to the user's real-time situation, so that the difficulty of the recommended exercise item is consistent with the current user's knowledge status, and in order to promptly know that the current user has mastered the current knowledge item, it is necessary to update the user's current knowledge status based on the item representation and learning feedback of the current exercise item.
[0140] Therefore, if the number of exercises in the current stage has not reached the current exercise limit, and the user's current knowledge status has not reached the state threshold, the low-level intelligent agent needs to select the next difficulty exercise item as the current exercise item according to the user's current knowledge status and the corresponding selection strategy for the user to continue learning, until the number of exercises in the current stage reaches the current exercise limit, that is, the user has not effectively mastered the current knowledge item and needs to re-determine the new current knowledge item, or the user's current knowledge status reaches the state threshold, that is, the user has mastered the current knowledge item and the next knowledge item can be recommended for rendering, and step S106 is executed to return to the high-level intelligent agent to determine the new current knowledge item.
[0141] Optionally, in another embodiment of the present application, a specific implementation of step S107 is as follows: Figure 4 As shown, the following steps are included:
[0142] S401. Select, through a low-level intelligent agent, from various practice items corresponding to the current knowledge item, a practice item with a difficulty equal to the current initial difficulty, as the current practice item, and provide it to the user for learning.
[0143] It should be noted that since the current initial difficulty is obtained by the high-level intelligent agent based on the analysis of the current knowledge state, its accuracy can be guaranteed. Therefore, in the embodiment of the present application, the low-level intelligent agent no longer performs further analysis, but directly selects the exercise items of this difficulty as the current exercise items and provides them to users for learning.
[0144] S402: Accumulate the number of exercises in the current stage and update the user's current knowledge status according to the project representation of the current exercise project and learning feedback.
[0145] S403: Determine whether the accumulated number of exercises in the current stage reaches the current exercise limit number, or whether the user's current knowledge status reaches the status threshold.
[0146] If it is determined that the accumulated number of exercises in the current stage does not reach the current exercise limit, and the user's current knowledge status does not reach the state threshold, step S404 is executed. If it is determined that the accumulated number of exercises in the current stage reaches the current exercise limit, or the user's current knowledge status reaches the state threshold, step S106 is executed.
[0147] S404: Connect the user's current knowledge state with the encoding vector of the current knowledge item to obtain the current state of the low-level intelligent agent.
[0148] That is, in the embodiment of the present application, the current state of the low-level agent includes the knowledge items c that the learner is currently learning. i and the learner’s knowledge state h at the current time t t . Similarly, the current knowledge item can also be represented by multi-hot encoding as , where k is the number of all knowledge items. Similarly, the position of the current knowledge item is 1, and the rest are 0. So the current state of the low-level agent can be represented as:
[0149]
[0150] in, Represents a concatenation operation.
[0151] S405. Using the actor-critic algorithm, based on the current state of the lower-level agent, a new current learning item is selected from the various practice items corresponding to the current knowledge item, and provided to the user for learning.
[0152] Specifically, the various practice items corresponding to the current knowledge item can be used as the practice action space, and the practice action is performed on the space, and a new current learning item is selected from it. Therefore, the practice action space can be expressed as:
[0153]
[0154] Among them, Ψ is a mapping function used to obtain the knowledge item c in the item hierarchy diagram. i A collection of associated exercise items.
[0155] In the embodiment of the present application, a critic algorithm framework similar to that of a high-level agent is used to obtain the selection strategy of a low-level agent to select the most appropriate exercise items for learners to practice. The main difference between the actor-critic algorithm and the proximal policy optimization algorithm is that the actor-critic algorithm uses a simple policy gradient method.
[0156] In order to control the difficulty of the current exercise item, the reward function of the low-level agent used in the embodiment of the present application is as follows:
[0157] r t p1 =L(DP t ,DP t-1 )=-(DP t -DP t-1 ) 2
[0158] Among them, DP t is the difficulty of the current exercise item at time t, DP t-1 It is the difficulty of the current exercise item at the last moment.
[0159] In addition, in order to evaluate the effectiveness of the current practice project, this scheme designs another part of the reward function calculation as follows:
[0160]
[0161] Among them, h i represents the knowledge state of the learner at learning stage i, and Thre is a predefined threshold.
[0162] Finally, the reward function of the low-level agent is obtained by the two balancing parameters α1 and α2, and the calculation is as follows:
[0163] r t p =α1×r t p1 +α2×r t p2 ; α1,α2∈[0,1]
[0164] It should be noted that after executing step S405, the process returns to executing step S402.
[0165] S108. The low-level agent uses the most recently updated number of exercises in the current stage as the number of exercises in the previous stage, and feeds back the most recently updated current knowledge state of the user to the high-level agent.
[0166] It should be noted that since the current stage of learning has ended, the user's current knowledge status and the number of exercises in the current stage need to be fed back to the high-level agent to start the next stage of learning. Therefore, after executing step S106, the execution returns to step S103.
[0167] To sum up, the framework of the learning path recommendation method provided in the embodiment of the present application can be implemented as follows: Figure 5 As shown in the figure. On the far left is the constructed project hierarchy diagram, and the corresponding representation is calculated based on the project hierarchy diagram. Figure 5 The right side of is the specific recommendation process, such as Figure 5 As shown, it is divided into multiple learning stages. For each learning stage, the high-level agent (L-Agent) determines a dynamic space from the layers of knowledge items in the project hierarchy diagram according to the user's knowledge status, and selects the current knowledge item from the dynamic space. At the same time, the current initial difficulty and the current practice limit will be determined and fed back to the low-level agent (P-Agent). Correspondingly, the low-level agent will select multiple practice items and recommend them to the user for learning, and update the user's knowledge status and cumulative learning times in combination with DIMKT. Finally, the low-level agent will feed back the user's knowledge status and cumulative learning times to the high-level agent to complete a stage of learning.
[0168] The present application implements a learning path recommendation method, which constructs a project hierarchical graph according to the association relationship between each learning project. Among them, the learning project includes multiple knowledge projects and their corresponding practice projects. The association relationship includes a predecessor-successor relationship and an inspection relationship. Then, the high-order information and difficulty information between the learning projects are aggregated through a hierarchical graph neural network to obtain the project representation of each practice project, so as to represent the practice project by integrating the representation of the difficulty information, so as to take into account the difficulty of the project when recommending the path. Then, the high-level intelligent agent determines the current knowledge project from the various target knowledge projects on the project hierarchical graph according to the user's current knowledge state and the number of exercises in the previous stage, and determines the current initial difficulty and the current practice limit number of times, and passes it to the low-level intelligent agent. Then, the low-level intelligent agent, starting from the current initial difficulty, gradually selects the current practice project of the corresponding difficulty from the various practice projects corresponding to the current knowledge project according to the user's current knowledge state, and provides it to the user for learning, and accumulates the number of exercises in the current stage and the project representation and learning feedback of the current practice project, and updates the user's current knowledge state until the number of exercises in the current stage reaches the current practice limit number of times, or the user's current knowledge state reaches the state threshold. The low-level agent uses the most recently updated number of exercises in the current stage as the number of exercises in the previous stage, and feeds back the most recently updated current knowledge state of the user to the high-level agent, so as to return to execute the high-level agent for the next stage of learning. Thus, through the high-level agent and the low-level agent, and the interaction between the two, a difficulty-driven method of recommending knowledge items to learn and exercise items under the knowledge items is implemented, thereby ensuring that the recommended learning path is smoother and improving learning efficiency.
[0169] Another embodiment of the present application provides a learning path recommendation device, such as Figure 6 As shown, including:
[0170] The construction module 601 is used to construct a project hierarchical diagram according to the association relationship between various learning projects.
[0171] Among them, the learning items include multiple knowledge items and their corresponding exercise items. The association relationship includes the predecessor-successor relationship and the inspection relationship.
[0172] Aggregation module 602 is used to aggregate high-level information and difficulty information between learning items through a hierarchical graph neural network to obtain item representations of various exercise items.
[0173] The high-level agent module 603 is used to determine the current knowledge item from the various target knowledge items on the project hierarchy diagram through the high-level agent according to the user's current knowledge status and the number of exercises in the previous stage, as well as the current initial difficulty and the current limit on the number of exercises, and pass them to the low-level agent.
[0174] The low-level intelligent agent module 604 is used to select current practice items of corresponding difficulty from various practice items corresponding to the current knowledge items through the low-level intelligent agent, starting from the current initial difficulty and gradually according to the current knowledge status of the user, and provide them to the user for learning, and accumulate the number of exercises in the current stage and update the current knowledge status of the user according to the item representation of the current practice items and learning feedback, until the number of exercises in the current stage reaches the current exercise limit, or the current knowledge status of the user reaches the status threshold.
[0175] The low-level agent uses the most recently updated number of exercises in the current stage as the number of exercises in the previous stage, and feeds back the most recently updated current knowledge status of the user to the high-level agent for return execution. The high-level agent determines the current knowledge item from each target knowledge item on the project hierarchy diagram according to the user's current knowledge status and the number of exercises in the previous stage, as well as the current initial difficulty and the current limit on the number of exercises until the user masters each target knowledge item.
[0176] Optionally, in the learning path recommendation device provided in another embodiment of the present application, the aggregation module includes:
[0177] The difficulty calculation module is used to calculate the difficulty of each learning item by multiplying the ratio of the number of incorrect answers to the total number of answers for each learning item by the corresponding preset difficulty level based on historical learning records.
[0178] The matrix generation module is used to generate a knowledge item embedding matrix and an exercise item embedding matrix, and to generate a knowledge item difficulty embedding matrix and an exercise item difficulty embedding matrix according to the difficulty of each learning item. The knowledge item embedding matrix and the exercise item embedding matrix are used to represent the full amount of knowledge items and exercise items, respectively.
[0179] A fusion module is used to, for each knowledge item, concatenate the item vector corresponding to the knowledge item in the knowledge item embedding matrix with the difficulty vector corresponding to the knowledge item in the knowledge item difficulty embedding matrix, multiply the result by the first weight matrix, and add the result with the first bias term to obtain the combined embedding vector of each knowledge item, and for each practice item, concatenate the item vector corresponding to the practice item in the practice item embedding matrix with the difficulty vector corresponding to the practice item in the practice item difficulty embedding matrix, multiply the result by the second weight matrix, and add the result with the second bias term to obtain the combined embedding vector of each practice item.
[0180] The combination module is used to connect the combined embedding vector of the exercise item with the mean of the combined embedding vectors of its adjacent knowledge items for each exercise item, so as to obtain the item representation of each exercise item.
[0181] Optionally, in a learning path recommendation device provided in another embodiment of the present application, when the high-level agent module is executed by the high-level agent to determine the current knowledge item from each target knowledge item on the item hierarchy diagram according to the user's current knowledge state and the number of exercises in the previous stage, and to determine the current initial difficulty and the current limit number of exercises, it is used to:
[0182] The current state of the high-level agent is obtained by connecting the user's current knowledge state with the encoding vectors of each target knowledge item through the high-level agent.
[0183] Using the proximal policy optimization algorithm, based on the current state of the high-level agent, the current knowledge item is determined from the current knowledge item dynamic space, where the current knowledge item dynamic space consists of the target knowledge items on the item hierarchical graph.
[0184] The current initial difficulty is calculated using the user's current knowledge status and the customized correct answer probability.
[0185] Through the trained multi-layer perceptron, the current practice limit is calculated based on the user's current knowledge status, the number of exercises in the previous stage, the practice limit in the previous stage, and the difficulty of the current knowledge item.
[0186] Optionally, in a learning path recommendation device provided in another embodiment of the present application, the low-level agent module executes the low-level agent to select current practice items of corresponding difficulty from various practice items corresponding to the current knowledge items step by step according to the current knowledge state of the user from the current initial difficulty, and provide them to the user for learning, and accumulate the number of exercises in the current stage and update the current knowledge state of the user according to the item representation and learning feedback of the current practice items, until the number of exercises in the current stage reaches the current exercise limit number, or the current knowledge state of the user reaches the state threshold, for:
[0187] A low-level intelligent agent selects an exercise item with a difficulty equal to the current initial difficulty from various exercise items corresponding to the current knowledge item, and provides it to the user for learning as the current exercise item.
[0188] The number of exercises in the current stage is accumulated and the user's current knowledge status is updated according to the project representation and learning feedback of the current exercise project.
[0189] Determine whether the accumulated number of exercises in the current stage has reached the current exercise limit, or whether the user's current knowledge status has reached the status threshold.
[0190] If it is determined that the accumulated number of exercises in the current stage has not reached the current exercise limit, and the user's current knowledge state has not reached the state threshold, the user's current knowledge state is connected with the encoding vector of the current knowledge item to obtain the current state of the low-level intelligent agent.
[0191] Using the actor-critic algorithm, based on the current state of the low-level intelligent agent, a new current learning item is selected from the various practice items corresponding to the current knowledge item and provided to the user for learning. The algorithm returns to accumulate the number of exercises in the current stage and updates the user's current knowledge state based on the item representation of the current practice item and learning feedback.
[0192] If it is determined that the accumulated number of exercises in the current stage reaches the current exercise limit, or the user's current knowledge status reaches the status threshold, the low-level agent will use the most recently updated number of exercises in the current stage as the number of exercises in the previous stage, and feed back the most recently updated current knowledge status of the user to the high-level agent.
[0193] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the implementation process of the corresponding steps in the above method embodiments, and will not be repeated here.
[0194] Another embodiment of the present application provides an electronic device, such as Figure 7 As shown, including:
[0195] Memory 701 and processor 702 .
[0196] The memory 701 is used to store programs.
[0197] The processor 702 is used to execute the program stored in the memory 701. When the program is executed, it is specifically used to implement the learning path recommendation method provided in any one of the above embodiments.
[0198] Another embodiment of the present application provides a computer storage medium for storing a computer program. When the computer program is executed, it is used to implement the learning path recommendation method provided in any one of the above embodiments.
[0199] Computer storage media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.
[0200] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0201] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A learning path recommendation method, characterized in that: include: According to the association relationship between various learning items, a project hierarchical diagram is constructed; wherein the learning items include multiple knowledge items and their corresponding exercise items; the association relationship includes a predecessor-successor relationship and an inspection relationship; Aggregating high-level information and difficulty information between the learning items through a hierarchical graph neural network to obtain item representations of each of the exercise items; The high-level agent determines the current knowledge item from each target knowledge item on the item hierarchy diagram according to the user's current knowledge state and the number of exercises in the previous stage, and determines the current initial difficulty and the current limit number of exercises, and transmits them to the low-level agent; The low-level agent gradually selects current practice items of corresponding difficulty from the practice items corresponding to the current knowledge items based on the current knowledge status of the user, starting from the current initial difficulty, and provides them to the user for learning, and accumulates the number of exercises in the current stage and updates the current knowledge status of the user based on the item representation and learning feedback of the current practice items, until the number of exercises in the current stage reaches the current exercise limit number, or the current knowledge status of the user reaches the status threshold; The low-level agent feeds back the most recently updated number of exercises in the current stage as the number of exercises in the previous stage and the most recently updated current knowledge status of the user to the high-level agent, so as to return and execute the high-level agent to determine the current knowledge item from the various target knowledge items on the project hierarchy diagram according to the user's current knowledge status and the number of exercises in the previous stage, as well as the current initial difficulty and the current limit on the number of exercises until the user masters each of the target knowledge items.
2. The method according to claim 1, characterized in that The high-level information and difficulty information between the learning items are aggregated through the hierarchical graph neural network to obtain the item representation of each of the exercise items, including: Based on the historical learning records, the ratio of the number of incorrect answers to the total number of answers for each learning item is multiplied by the corresponding preset difficulty level to obtain the difficulty of each learning item; Generate a knowledge item embedding matrix and an exercise item embedding matrix, and generate a knowledge item difficulty embedding matrix and an exercise item difficulty embedding matrix according to the difficulty of each of the learning items; wherein the knowledge item embedding matrix and the exercise item embedding matrix are respectively used to represent the full amount of the knowledge items and the exercise items; For each of the knowledge items, respectively, the item vector corresponding to the knowledge item in the knowledge item embedding matrix and the difficulty vector corresponding to the knowledge item in the knowledge item difficulty embedding matrix are concatenated, multiplied by a first weight matrix, and added with a first bias term to obtain a combined embedding vector of each of the knowledge items; and for each of the practice items, the item vector corresponding to the practice item in the practice item embedding matrix and the difficulty vector corresponding to the practice item in the practice item difficulty embedding matrix are concatenated, multiplied by a second weight matrix, and added with a second bias term to obtain a combined embedding vector of each of the practice items; For each of the exercise items, the combined embedding vector of the exercise item is connected with the mean of the combined embedding vectors of its adjacent knowledge items to obtain the item representation of each of the exercise items.
3. The method according to claim 1, characterized in that The high-level agent determines the current knowledge item from each target knowledge item on the item hierarchy diagram according to the user's current knowledge state and the number of exercises in the previous stage, and determines the current initial difficulty and the current number of exercise limits, including: The high-level agent connects the current knowledge state of the user with the encoding vectors of each target knowledge item to obtain the current state of the high-level agent; Determine the current knowledge item from the current knowledge item dynamic space based on the current state of the high-level agent using a proximal strategy optimization algorithm; wherein the current knowledge item dynamic space is composed of the target knowledge item on the item hierarchical graph; The current initial difficulty is calculated using the user's current knowledge status and a custom correct answer probability; The current practice limit number of times is calculated by a trained multi-layer perceptron based on the user's current knowledge status, the practice number of the previous stage, the practice limit number of the previous stage, and the difficulty of the current knowledge item.
4. The method according to claim 1, characterized in that: The low-level agent gradually selects current practice items of corresponding difficulty from the practice items corresponding to the current knowledge items according to the current knowledge state of the user, starting from the current initial difficulty, and provides them to the user for learning, and accumulates the number of exercises in the current stage and updates the current knowledge state of the user according to the item representation and learning feedback of the current practice items, until the number of exercises in the current stage reaches the current exercise limit number, or the current knowledge state of the user reaches the state threshold, including: Selecting, by the low-level agent, from the various exercise items corresponding to the current knowledge item, an exercise item with a difficulty of the current initial difficulty as the current exercise item, and providing it to the user for learning; Accumulating the number of exercises in the current stage and updating the current knowledge state of the user according to the project representation and learning feedback of the current exercise project; Determine whether the accumulated number of exercises in the current stage reaches the current exercise limit number, or whether the current knowledge status of the user reaches a status threshold; If it is determined that the accumulated number of exercises in the current stage does not reach the current exercise limit number, and the current knowledge state of the user does not reach the state threshold, the current knowledge state of the user is connected with the encoding vector of the current knowledge item to obtain the current state of the low-level agent; Using the actor-critic algorithm, based on the current state of the low-level agent, a new current learning item is selected from the various practice items corresponding to the current knowledge item, provided to the user for learning, and the current knowledge state of the user is updated according to the item representation and learning feedback of the current practice item and the accumulated number of practice times in the current stage; If it is determined that the accumulated number of exercises in the current stage reaches the current exercise limit, or the current knowledge status of the user reaches the status threshold, the low-level agent will use the currently updated number of exercises in the current stage as the number of exercises in the previous stage, and feed back the currently updated number of exercises in the current stage and the current knowledge status of the user to the high-level agent.
5. A learning path recommendation device, characterized in that: include: A construction module is used to construct a project hierarchical diagram according to the association relationship between various learning projects; wherein the learning projects include multiple knowledge projects and their corresponding exercise projects; the association relationship includes a predecessor-successor relationship and an inspection relationship; An aggregation module, used to aggregate high-level information and difficulty information between the learning items through a hierarchical graph neural network to obtain item representations of each of the exercise items; A high-level agent module, used to determine the current knowledge item from each target knowledge item on the item hierarchy diagram according to the user's current knowledge state and the number of exercises in the previous stage through the high-level agent, as well as the current initial difficulty and the current limit number of exercises, and pass them to the low-level agent; A low-level agent module is used to select, from the current initial difficulty, through the low-level agent, a current exercise item of corresponding difficulty from each exercise item corresponding to the current knowledge item according to the current knowledge state of the user, and provide it to the user for learning, and accumulate the number of exercises in the current stage and update the current knowledge state of the user according to the item representation and learning feedback of the current exercise item, until the number of exercises in the current stage reaches the current exercise limit number, or the current knowledge state of the user reaches a state threshold; The low-level agent feeds back the most recently updated number of exercises in the current stage as the number of exercises in the previous stage and the most recently updated current knowledge status of the user to the high-level agent, so as to return and execute the high-level agent to determine the current knowledge item from the various target knowledge items on the project hierarchy diagram according to the user's current knowledge status and the number of exercises in the previous stage, as well as the current initial difficulty and the current limit on the number of exercises until the user masters each of the target knowledge items.
6. The device according to claim 5, characterized in that The aggregation module comprises: A difficulty calculation module, for obtaining the difficulty of each learning item by multiplying the ratio of the number of incorrect answers to the total number of answers for each learning item by the corresponding preset difficulty level based on historical learning records; A matrix generation module, used to generate a knowledge item embedding matrix and an exercise item embedding matrix, and to generate a knowledge item difficulty embedding matrix and an exercise item difficulty embedding matrix according to the difficulty of each of the learning items; wherein the knowledge item embedding matrix and the exercise item embedding matrix are respectively used to represent the full amount of the knowledge items and the exercise items; A fusion module, for respectively, for each of the knowledge items, concatenating the item vector corresponding to the knowledge item in the knowledge item embedding matrix with the difficulty vector corresponding to the knowledge item in the knowledge item difficulty embedding matrix, multiplying the result of the concatenation operation by a first weight matrix, and adding the result with a first bias term, to obtain a combined embedding vector of each of the knowledge items; and for respectively, for each of the practice items, concatenating the item vector corresponding to the practice item in the practice item embedding matrix with the difficulty vector corresponding to the practice item in the practice item difficulty embedding matrix, multiplying the result of the concatenation operation by a second weight matrix, and adding the result with a second bias term, to obtain a combined embedding vector of each of the practice items; The combining module is used to connect the combined embedding vector of each exercise item with the mean of the combined embedding vectors of its adjacent knowledge items for each exercise item, so as to obtain the item representation of each exercise item.
7. The device according to claim 5, characterized in that When the high-level agent module executes the step of determining the current knowledge item from each target knowledge item on the item hierarchy diagram according to the user's current knowledge state and the number of exercises in the previous stage through the high-level agent, and determining the current initial difficulty and the current limit number of exercises, it is used to: The high-level agent connects the current knowledge state of the user with the encoding vectors of each target knowledge item to obtain the current state of the high-level agent; Determine the current knowledge item from the current knowledge item dynamic space based on the current state of the high-level agent using a proximal strategy optimization algorithm; wherein the current knowledge item dynamic space is composed of the target knowledge item on the item hierarchical graph; The current initial difficulty is calculated using the user's current knowledge status and a custom correct answer probability; The current practice limit number of times is calculated by a trained multi-layer perceptron based on the user's current knowledge status, the practice number of the previous stage, the practice limit number of the previous stage, and the difficulty of the current knowledge item.
8. The device according to claim 5, characterized in that The low-level agent module executes the step of selecting current practice items of corresponding difficulty from the practice items corresponding to the current knowledge items through the low-level agent starting from the current initial difficulty, step by step according to the current knowledge state of the user, providing the current practice items with corresponding difficulty to the user for learning, accumulating the number of practice times in the current stage, and updating the current knowledge state of the user according to the item representation and learning feedback of the current practice items, until the number of practice times in the current stage reaches the current practice limit number, or the current knowledge state of the user reaches the state threshold, and is used to: Selecting, by the low-level agent, from the various exercise items corresponding to the current knowledge item, an exercise item with a difficulty of the current initial difficulty as the current exercise item, and providing it to the user for learning; Accumulating the number of exercises in the current stage and updating the current knowledge state of the user according to the project representation and learning feedback of the current exercise project; Determine whether the accumulated number of exercises in the current stage reaches the current exercise limit number, or whether the current knowledge status of the user reaches a status threshold; If it is determined that the accumulated number of exercises in the current stage does not reach the current exercise limit number, and the current knowledge state of the user does not reach the state threshold, the current knowledge state of the user is connected with the encoding vector of the current knowledge item to obtain the current state of the low-level agent; Using the actor-critic algorithm, based on the current state of the low-level agent, a new current learning item is selected from the various practice items corresponding to the current knowledge item, provided to the user for learning, and the current knowledge state of the user is updated according to the item representation and learning feedback of the current practice item and the accumulated number of practice times in the current stage; If it is determined that the accumulated number of exercises in the current stage reaches the current exercise limit, or the current knowledge status of the user reaches the status threshold, the low-level agent will use the currently updated number of exercises in the current stage as the number of exercises in the previous stage, and feed back the currently updated number of exercises in the current stage and the current knowledge status of the user to the high-level agent.
9. An electronic device, characterized in that: include: Memory and processor; Wherein, the memory is used to store programs; The processor is used to execute the program, and when the program is executed, it is specifically used to implement the learning path recommendation method according to any one of claims 1 to 4.
10. A computer storage medium, characterized in that: Used to store a computer program, which, when executed, is used to implement the learning path recommendation method as described in any one of claims 1 to 4.
Citation Information
Patent Citations
Learning path recommendation method and system combining graph neural network and multi-layer perceptron
CN117094859A
Knowledge graph recommendation method and system based on improved KGAT model
WO2023097929A1
