An education resource recommendation method based on human-computer bidirectional interaction
By constructing encoded information of learners and learning resources and feedback optimization algorithms, and combining cloud-edge collaborative computing and graph neural networks, the problems of poor personalized recommendation effect and catastrophic forgetting in educational resource recommendation systems are solved, and efficient and personalized educational resource recommendation services are realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUAZHONG NORMAL UNIV
- Filing Date
- 2025-06-30
- Publication Date
- 2026-07-31
AI Technical Summary
Existing educational resource recommendation systems struggle to efficiently and dynamically adapt to changes in learners' needs, resulting in poor personalized recommendations and a catastrophic forgetting problem.
By constructing encoded information about learners and learning resources, combining feedback optimization algorithms and real-time user interaction, utilizing cloud-edge collaborative computing, continuously updating recommendations, and employing graph neural networks to analyze matching degree and optimize encoded information, a personalized educational resource recommendation service is constructed.
This system enables the educational resource recommendation system to efficiently and dynamically adapt to learners' needs, ensuring that recommended resources align with learners' interests, thus improving the recommendation experience and satisfaction. It also overcomes the problems of insufficient human-computer trust and poor interpretability in traditional methods.
Smart Images
Figure CN120929507B_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of educational informatization and intelligent recommendation technology, and more specifically, relates to an educational resource recommendation method based on human-computer two-way interaction. Background Technology
[0002] In the field of modern recommender systems, the focus of technological development has gradually shifted from simply pursuing the accuracy of recommendation results to a deeper concern for the overall user experience. To improve user satisfaction and engagement, many applications have designed and deployed diverse user interaction channels to capture user preferences, feedback, and behavioral patterns. Especially in the specific application scenario of educational resource recommendation, given the dynamic and personalized nature of users' learning needs, knowledge levels, and interests, how to efficiently update recommendations is a pressing technical problem to be solved in this field. Summary of the Invention
[0003] To address the shortcomings of existing technologies, this application aims to provide an educational resource recommendation method based on two-way human-computer interaction. This method constructs encoded information about learners and learning resources based on resource knowledge graphs and learners' historical learning records. Through a proposed feedback optimization algorithm, it interacts with users in real time, efficiently updating recommendations. It allows users to provide real-time feedback and effectively responds to improve the recommendation results. This method has wide applicability and offers a high level of recommendation experience and satisfaction.
[0004] To achieve the above objectives, firstly, this application provides a method for recommending educational resources based on two-way human-computer interaction, the method comprising: Based on the learner's coding information and the coding information of the learning resources, the matching degree between the learner and the learning resources is analyzed to determine the initial recommended learning resources and corresponding keywords. The learner's coding information is used to characterize the learner's characteristics, the learning resource's coding information is used to characterize the learning resource's characteristics, and the keywords are used to represent the key information of the learning resources. Based on the keywords provided by learners, the learners' coding information is continuously updated, and the recommended learning resources and corresponding keywords are determined based on the updated learners' coding information, until the learners no longer provide keywords. The keywords for learner feedback are determined by displaying keywords corresponding to recommended learning resources and receiving the learner's target input, which is used to select one or more keywords as feedback keywords from the displayed keywords.
[0005] Understandably, the system analyzes the matching degree based on the learner's and resource's coding information to generate initial recommendations. Then, by displaying keywords of the recommended resources and receiving feedback from learners, it continuously adjusts the learner's coding information to optimize the recommendation results. This two-way interactive process enables the recommendation system to dynamically adapt to changes in learners' needs, ensuring that recommended resources remain highly consistent with the learner's current learning status and interests, until the learner no longer provides new feedback.
[0006] Therefore, the method provided in this application achieves efficient recommendation updates by combining the encoded information of learners and learning resources with a human-computer interaction feedback mechanism.
[0007] Optionally, in a cloud-edge collaborative computing environment, the method provided in this application first utilizes the powerful computing capabilities of the cloud to analyze the matching degree between the learner's coding information and the coding information of the learning resources, determining the initial recommended learning resources and their corresponding keywords. The learner's coding information characterizes the learner's traits, the coding information of the learning resources reflects the resource's characteristics, and the keywords intuitively present key information about the resources. Subsequently, at the edge, the system displays the recommended learning resource keywords to the learner, receiving their target input—that is, the learner selects one or more keywords from the displayed keywords as feedback and reports it to the cloud. Based on this feedback, the cloud quickly updates the learner's coding information and the recommended learning resources, and synchronizes the updated data to the edge. This cloud-edge collaboration continues iteratively until the learner no longer provides keywords, thereby providing learners with a highly personalized and accurately tailored educational resource recommendation service.
[0008] Optionally, a knowledge graph of learning resources can be constructed based on various related information of the learning resources (including keywords of the learning resources). By searching the knowledge graph of learning resources, the keywords corresponding to the learning resources can be obtained.
[0009] In one possible implementation, the continuous updating of learners' encoding information and the determination of recommended learning resources and corresponding keywords based on the updated learners' encoding information include: Based on the keywords provided by learners, representative items are sampled from the learning resource knowledge graph. The learning resource knowledge graph is constructed based on various related information of learning resources (including keywords of learning resources). The representative items are learning resource entities in the learning resource knowledge graph that have a connection relationship with the keywords provided by learners. Based on the sensitivity of learners' encoded information to perturbations, a gradient-based regularizer is constructed to control the updating of learners' encoded information. Based on the encoding information of representative projects and regularizers, the learner's encoding information is updated through gradient descent. Based on the encoding information of learning resources and the updated learner's encoding information, the latest recommended learning resources and corresponding keywords are determined by analyzing the matching degree between learners and learning resources.
[0010] This section elaborates on representative items. To enhance the utilization of keywords, after learners provide keywords, the system searches the learning resource knowledge graph for learning resource entities directly or indirectly connected to these keywords. These entities constitute the representative items. The connections between nodes (such as keywords and resources) in the knowledge graph reflect their relevance in content or topic. Therefore, learning resource entities connected to the feedback keywords are likely also related to the learner's current interests or needs in terms of content. These representative items not only contain user feedback preference information but also rich collaborative information (collaborative information refers to the group preferences of a specific user group (the user's group) revealed by the interrelationships between representative items). Incorporating these representative items allows for a more accurate capture of the learner's new intentions or preferences expressed through feedback keywords when updating learner coding information and generating new recommendations, thus making the recommendations more aligned with the learner's latest needs.
[0011] In one possible implementation, the aforementioned keywords based on learner feedback are used to sample representative items in the learning resource knowledge graph, including sampling representative items using the following formula: ; in, This represents the set of representative items sampled. This indicates a multi-order sampling method, where the multi-order sampling method is proportional. Sampling learning resource knowledge graph Keywords in learner feedback Representative projects at different levels with interconnected relationships are set up. Indicates from the first The proportion of representative items in the middle of the sampling to the total number of samples.
[0012] In one possible implementation, the above regularizer is constructed using the following formula: ; ; ; in, This represents the prior probability of the learner's encoded information. Indicates weight, This represents the initial encoded information of the learner. This represents the updated learner's coding information. Indicates regularization, This represents all the learning resources that the learner has interacted with in their historical learning record. This represents the total number of all learning resources that a learner has interacted with in their historical learning record. express The The sensitivity of each dimension to disturbances Indicates learner, This represents the distribution matched with learners' historical learning records. The feature vector of the learner's encoded information. express The Each dimension.
[0013] Understandably, to mitigate the catastrophic forgetting problem arising from the updating of user-encoded information, a gradient-based regularizer is constructed. This controls the updating of user-encoded information and strengthens the algorithm's memory of important user preference features.
[0014] In one possible implementation, the learner's encoded information is updated via gradient descent based on the encoded information representing the item and the regularizer, including: By using gradient descent, the loss function is minimized, and the learner's encoded information is updated. ; in, Indicates the loss value. The feedback from learners is characterized as positive or negative. Indicates the total number of samples. This represents the feedback collected from learners. Indicates learner, Indicates the learner's ID number. Keywords representing learner feedback Indicates the keyword number. It is a constant. This represents the sigmoid function. This represents a feature vector that encodes information about the project. This indicates the project number. This represents the preset hyperparameters.
[0015] In one possible implementation, the learner's encoded information and the encoded information of the learning resources are obtained through the following steps: The system acquires a knowledge graph of learning resources and learners' historical learning records. The knowledge graph of learning resources is constructed based on various related information of learning resources (including keywords of learning resources). Based on the learning resources in the learner's historical learning records, the corresponding learning resource entities are found in the learning resource knowledge graph; Based on the learning resource entities corresponding to the learning resources in the learner's historical learning records, the learner's historical learning records are spliced with the learning resource knowledge graph to construct a collaborative knowledge graph; Based on collaborative knowledge graphs, graph neural networks are used to obtain the encoded information of learners and the encoded information of learning resources.
[0016] Understandably, the system constructs a knowledge graph of learning resources that includes the relationships between learning resources. By utilizing learners' historical learning records, it locates specific learning resource entities in the graph. Then, it splices the learners' learning trajectories with the knowledge graph to form a collaborative knowledge graph that can simultaneously reflect learners' behavior and resource relationships. Furthermore, by using graph neural networks to perform in-depth analysis of the collaborative knowledge graph, it can capture the complex relationships between learning resources and the interaction patterns between learners and these resources, thereby extracting encoded information that can represent learners' characteristics (such as interests and abilities) and resource characteristics (such as content and difficulty).
[0017] In one possible implementation, the above analysis of the match between learners and learning resources includes: Based on the learner's encoded information, a feature vector of the learner's encoded information is generated through a graph neural network; Based on the encoded information of the learning resources, a feature vector of the encoded information of the learning resources is generated through a graph neural network. The matching degree between learners and learning resources is determined based on the feature vectors of the learners' encoded information and the feature vectors of the learning resources' encoded information.
[0018] In one possible implementation, recommended learning resources are determined through the following steps: Sort the learning resources from highest to lowest matching degree; Select the first in sorting These learning resources are recommended learning resources. It is a positive integer.
[0019] In a second aspect, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.
[0020] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.
[0021] It is understood that the beneficial effects of the second and third aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.
[0022] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: The system analyzes the matching degree based on the learner's and resource's coding information to generate initial recommendations. Then, by displaying keywords of the recommended resources and receiving feedback from learners, it continuously adjusts the learner's coding information to optimize the recommendation results. This two-way interactive process allows the recommendation system to dynamically adapt to changes in learners' needs, ensuring that recommended resources remain highly consistent with the learner's current learning status and interests until the learner provides no further feedback. Therefore, the method provided in this application achieves efficient recommendation updates by combining learner and learning resource coding information with a human-computer interaction feedback mechanism. Attached Figure Description
[0023] Figure 1 This is a flowchart illustrating the educational resource recommendation method based on two-way human-computer interaction provided in an embodiment of this application. Figure 2 This is a schematic diagram of the feedback optimization algorithm provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation
[0024] To facilitate a clearer understanding of the various embodiments of this application, some relevant background knowledge will be introduced as follows.
[0025] Recent research on human-computer interaction in recommender systems has focused on conversational recommendation scenarios. These methods typically involve two steps: first, in the training phase, designing a base model based on the user's historical records to train initial recommendations; and second, in the intervention phase, analyzing user preferences and gradually adjusting recommendations based on positive or negative feedback from the user during interactions with the system. Because keywords provide more precise semantic information, most research has focused on utilizing keywords in human-computer interaction.
[0026] Although these methods demonstrate the ability to adjust recommendations based on user interaction, three problems remain to be solved. (1) In order to accept user feedback during the intervention phase, these works have designed special models during the training phase that can directly model the relationship between users and keywords. Although this enables the model to accept user feedback, it greatly limits the applicable recommendation scenarios and the initial recommendation performance is poor; (2) The collaborative information of feedback is generally ignored, there is a lack of adjustment for items with similar content, and the utilization of keywords is low; (3) Ignoring the intervention phase, the cumulative changes in continuous parameter adjustments reduce the model's ability to capture users' old preferences, leading to catastrophic forgetting.
[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0028] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.
[0029] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.
[0030] The embodiments of this application are described below with reference to the accompanying drawings.
[0031] Figure 1 This is a flowchart illustrating the educational resource recommendation method based on human-computer two-way interaction provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes the following steps.
[0032] Step (1) Extract the encoding information of learners and learning resources from the encoding information database. Before recommending educational resources, encoding information can be generated for different learners and learning resources. For example, the learner encoding space is the total number of learners N multiplied by the dimension of the encoding information (taking a 64-dimensional vector as an example). (a dimensional vector space), and store the generated encoded information in the encoded information database.
[0033] Step (2) Based on the learner and learning resource coding information, the matching degree between learners and learning resources is obtained. A ranking is generated from high to low according to the matching degree score of the project (i.e., the learning resource). Initial recommended projects (i.e., the initial recommended learning resources) are generated based on the ranking, and keywords for the recommended projects are provided based on the knowledge graph. For example, when recommending videos related to "Lists" in the "Data Structures" course, video-related keyword information, such as video source and instructor information, is provided to the user as project keywords. Step (3) Collect feedback keywords selected by learners. For example, keywords of the instructors in the course videos above, and update the learner coding information in the coding information database.
[0034] Step (4) Multiply the learner's feature encoding vector with the learning resource's feature encoding vector again to obtain the matching degree between the learner and the learning resource, and generate an optimized recommendation.
[0035] Step (5) Repeat steps (3) and (4) above until the user no longer provides new feedback.
[0036] Furthermore, the coding information database in step (1) is determined as follows: This process aggregates various information about learners and learning resources, such as a knowledge graph composed of learners' historical learning records and various related information about learning resources. Based on the learning resources in the historical learning records, the corresponding learning resource entities are found in the knowledge graph and aligned. Then, the knowledge graphs of historical learning records (bipartite graph) and learning resources are stitched together into a new graph (called a collaborative knowledge graph). Based on the collaborative knowledge graph, the encoded information of learners and learning resources is obtained through a graph neural network.
[0037] Furthermore, in step (2), the learner-learner matching degree and project keywords are determined as follows: use and Let represent the feature vectors of the learner's encoded information and the feature vectors of the learning resources' encoded information, respectively, generated by the graph neural network. and These represent the encoded information for the corresponding learner and learning resource, respectively. Indicates the learner's ID number. This indicates the number of the learning resource. and These represent the parameters in the graph neural network. The form can be constructed using different graph neural networks, such as graph convolutional neural networks and graph attention neural networks.
[0038] After obtaining the feature vectors of the learner and learning resources output by the graph neural network, the corresponding matching degree is calculated by the following formula: ; At the same time, the system will provide an educational resource knowledge graph. China's Recommended Projects Keywords This serves as a recommendation explanation and guides user interaction.
[0039] Furthermore, the learner coding information updated in step (3) is determined as follows: use This represents the collected user feedback, among which This refers to the user (i.e., the learner). The updated learner coding information can be calculated using the following formula, which represents the feedback keywords provided by the user: ; in These represent the encoding information of learning resources and the encoding information of feedback keywords, respectively. This represents the posterior probability of user-encoded information obtained from a given user response. Indicates learner's... The probability of preference for Chinese keywords. Encode prior probabilities of information for users.
[0040] Furthermore, in step (4), the matching degree between the learner and the learning resources in step (2) is recalculated to obtain an optimized recommendation. Steps (3) and (4) above are repeated until the user stops interacting with the system (no longer providing keywords).
[0041] In one possible implementation, updating the learner's encoding information in the encoding information base and generating optimized recommendations can specifically employ a feedback optimization algorithm.
[0042] Figure 2 This is a schematic diagram of the feedback optimization algorithm provided in the embodiments of this application, such as... Figure 2 As shown, the algorithm includes the following modules: The first module samples representative items in the knowledge graph based on keywords fed back by users (i.e. learners) as concrete preferences of user feedback, and updates user coding information based on representative items. The second module, to mitigate the catastrophic forgetting problem arising during the user's encoded information update process, constructs a gradient-based regularizer. Controlling the updating of user coding information and strengthening the algorithm's memory of important user preference features; The third module, based on the encoding information of representative projects and a regularizer, updates the user encoding information through gradient descent. Then, the updated user encoding information and the learning resource encoding information are input into a graph neural network to obtain the updated learner-to-learning resource matching score. The recommendation is updated based on the matching score. Furthermore, based on the updated user encoding information and the feature encoding of the learning resources, the updated learner-to-learning resource matching degree is obtained, and the recommendation is updated based on the matching degree.
[0043] Specifically, the algorithm optimizes the posterior of user-encoded information by calculating the probability of user preference for feedback keywords, while preserving the prior of user preferences as much as possible. Based on maximum a posteriori estimation, the algorithm's optimization objective is to maximize the posterior estimate of user-encoded information: .
[0044] Furthermore, the first module includes: First of all, for This can indicate the learner's understanding of... Preference for Chinese keywords Direct calculation This is difficult, so we indirectly compute by sampling representative items from the knowledge graph. Representative items refer to those highly relevant to user feedback keywords obtained through sampling. These items not only contain user feedback preference information but also rich collaborative information (collaborative information refers to the group preferences of a specific user group (the user's group) revealed by the correlations between representative items). Formally, Converted into preference scores between learners and representative items ( (representing the project) and the probability of representing the project and keywords. The product of: ; Expressing expectations, for If paths can be observed between them in the educational knowledge graph, then... Set it to 1 otherwise to 0, and use the Monte Carlo method to approximate the above formula: ; in This represents the representative items obtained based on the sampling method. This represents the set of representative items sampled. This refers to the total number of samples (the total number of representative items obtained from the sampling). A multi-level sampling method is used to sample representative items related to keywords from the knowledge graph. ; in, This represents a multi-level sampling method, using a proportionally sampled knowledge graph. Middle and feedback keywords Representative projects at different levels with interconnected relationships are set up. Indicates from the first The proportion of items sampled in the middle of the total number of samples The proportion. In particular, the sampling method is set to random sampling.
[0045] Furthermore, the second module includes: For a given group of learners' historical learning records ( Indicates learner, (Representing learning resources), used This indicates the distribution of its matches. This represents the feature vector of the user's encoded information, where... This represents a single dimension in the encoded information vector. Then, for... generate disturbance The change in the final predicted value is represented as follows: ; Then it can be calculated The magnitude of the vector weights measures their importance, i.e., how much a perturbation to this parameter will change the current prediction. Finally, the importance weights of each user on the parameter are calculated by summing them up. ; in, Indicates regularization, It refers to all the learning resources that the user (learner) has interacted with in their historical learning records. This represents the total number of all learning resources the user (learner) has interacted with in their historical learning record (interaction can be the learner accessing learning resources). Changes to parameters with lower importance weights will not have a significant impact on the output; therefore, changes to these parameters will not be subject to many restrictions in subsequent tasks. Changes to parameters with higher importance weights, however, will be constrained or ideally should remain unchanged. ; in This represents the updated user-encoded information vector. This represents the initial user-coded information vector. This indicates that the encoded information vector of the user (learner) is summed to facilitate backpropagation and gradient calculation.
[0046] Furthermore, the third module includes: First, the preference scores between learners and proxy projects. Given by the following formula: ; in It is the feature vector of the user's encoded information. It is a feature vector representing the encoded information of the project. It is a constant. This represents the sigmoid function. Finally, it is given by the following formula: ; in This indicates that user feedback is positive. This represents negative feedback. Ultimately, the maximum posterior probability of the user feature encoding vector can be transformed into minimizing the loss function, as follows: ; in, These are the preset hyperparameters.
[0047] Understandably, compared to traditional recommendation methods, this application can utilize educational knowledge graph data to obtain user feedback in a timely manner, continuously execute feedback optimization algorithms, and construct a personalized educational resource recommendation service that is controllable and understandable to users. Experiments show high user satisfaction, overcoming problems such as insufficient human-computer trust and poor interpretability in traditional recommendation methods.
[0048] Based on the methods in the above embodiments, this application provides an electronic device. Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application, such as... Figure 3 As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute the methods in the above embodiments.
[0049] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.
[0050] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.
[0051] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.
[0052] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.
[0053] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.
[0054] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).
[0055] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.
[0056] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for recommending educational resources based on bidirectional human-computer interaction, characterized in that, include: Based on the learner's coding information and the coding information of the learning resources, the matching degree between the learner and the learning resources is analyzed to determine the initial recommended learning resources and corresponding keywords. The learner's coding information is used to characterize the learner's characteristics, the learning resource's coding information is used to characterize the learning resource's characteristics, and the keywords are used to represent the key information of the learning resources. Based on the keywords provided by learners, the learners' coding information is continuously updated, and the recommended learning resources and corresponding keywords are determined based on the updated learners' coding information, until the learners no longer provide keywords. The keywords for learner feedback are determined by displaying keywords corresponding to recommended learning resources and receiving the learner's target input, which is used to select one or more keywords as feedback keywords from the displayed keywords. The process of continuously updating learners' coding information and determining recommended learning resources and corresponding keywords based on the updated learners' coding information includes: Based on the keywords provided by learners, representative items are sampled from the learning resource knowledge graph. The learning resource knowledge graph is constructed based on various related information of learning resources. The representative items are the learning resource entities in the learning resource knowledge graph that have a connection relationship with the keywords provided by learners. Based on the sensitivity of learners' encoded information to perturbations, a gradient-based regularizer is constructed to control the updating of learners' encoded information. Based on the encoding information of representative projects and regularizers, the learner's encoding information is updated through gradient descent. Based on the encoding information of learning resources and the updated encoding information of learners, the latest recommended learning resources and corresponding keywords are determined by analyzing the matching degree between learners and learning resources. The sampling of representative items in the learning resource knowledge graph based on learner feedback keywords includes sampling representative items using the following formula: ； wherein, represents a set of sampled representative items, represents a multi-order sampling method, the multi-order sampling method proportionally sampling a learning resource knowledge graph key words in the learning resource and the feedback of the learner different orders of representative items with a connected relationship are set represents a proportion of the number of representative items sampled from the order to the total number of samples; The regularizer is constructed using the following formula: ； ； ； in, This represents the prior probability of the learner's encoded information. Indicates weight, This represents the initial encoded information of the learner. This represents the updated learner's coding information. Indicates regularization, This represents all the learning resources that the learner has interacted with in their historical learning record. This represents the total number of all learning resources that a learner has interacted with in their historical learning record. express The The sensitivity of each dimension to disturbances Indicates learner, This represents the distribution matched with learners' historical learning records. The feature vector of the learner's encoded information. express The One dimension; The process of updating the learner's encoding information via gradient descent based on the representative item's encoding information and a regularizer includes: By using gradient descent, the loss function is minimized, and the learner's encoded information is updated. ； in, Indicates the loss value. The feedback from learners is characterized as positive or negative. Indicates the total number of samples. This represents the feedback collected from learners. Indicates learner, Indicates the learner's ID number. Keywords representing learner feedback Indicates the keyword number. It is a constant. This represents the sigmoid function. This represents a feature vector that encodes information about the project. This indicates the project number. This represents the preset hyperparameters.
2. The educational resource recommendation method based on human-computer two-way interaction according to claim 1, characterized in that, The learner's coding information and the coding information of the learning resources are obtained through the following steps: Obtain a knowledge graph of learning resources and learners' historical learning records. The knowledge graph of learning resources is constructed based on various related information of learning resources. Based on the learning resources in the learner's historical learning records, the corresponding learning resource entities are found in the learning resource knowledge graph; Based on the learning resource entities corresponding to the learning resources in the learner's historical learning records, the learner's historical learning records are spliced with the learning resource knowledge graph to construct a collaborative knowledge graph; Based on collaborative knowledge graphs, graph neural networks are used to obtain the encoded information of learners and the encoded information of learning resources.
3. The educational resource recommendation method based on human-computer two-way interaction according to any one of claims 1-2, characterized in that, The analysis of the match between learners and learning resources includes: Based on the learner's encoded information, a feature vector of the learner's encoded information is generated through a graph neural network; Based on the encoded information of the learning resources, a feature vector of the encoded information of the learning resources is generated through a graph neural network. The matching degree between learners and learning resources is determined based on the feature vectors of the learners' encoded information and the feature vectors of the learning resources' encoded information.
4. The educational resource recommendation method based on human-computer two-way interaction according to claim 3, characterized in that, Recommended learning resources were determined through the following steps: Sort the learning resources from highest to lowest matching degree; Select the first in sorting These learning resources are recommended learning resources. It is a positive integer.
5. An electronic device, characterized in that, include: At least one memory for storing computer programs; At least one processor is configured to execute a program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to perform the method as described in any one of claims 1-4.
6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is run on the processor, it causes the processor to perform the method as described in any one of claims 1-4.