A knowledge graph-based learning resource recommendation method

Abstract

The embodiment of the application discloses a learning resource recommendation method based on a knowledge graph, relates to the technical field of intelligent teaching, and comprises the following steps: acquiring a learner portrait of a learner and various types of learning resources used by the learner, and generating a multi-modal learning resource interaction graph; calculating a contrast loss function based on the multi-modal learning resource interaction graph; acquiring a matching coefficient of the learner and the learning resources based on the interaction relationship between the learner and the learning resources; calculating the interest degree of the learner to various types of learning resources based on the numerical value of the matching coefficient; sorting various types of learning resources based on the interest degree, and recommending learning resources to the corresponding learner in sequence based on the sorting result. By adopting the method, multi-modal information fusion of learning resources and historical learning data of learners are considered, and the two are jointly integrated into a learning resource interaction network, so that the expression meaning is enriched, and the precision of learning resource adaptation is improved.

Description

Technical Field

[0001] This application relates to the field of intelligent teaching technology, and in particular to a method, apparatus, device and storage medium for recommending learning resources based on knowledge graphs. Background Technology

[0002] Current online education platforms can recommend learning resources to students through recommendation systems, thereby enabling learners to better acquire knowledge and learn from resources that suit them.

[0003] Research shows that learners are overwhelmed by the sheer volume of information available when choosing a resource, and the range of course-related information available to learners is rapidly increasing. This abundance of information makes it more difficult for students to select, organize, and use resources that align with their personal goals and interests.

[0004] On the one hand, adapting rich learning resources to learners has become a huge challenge; on the other hand, traditional learning resource recommendation and guidance programs often rely on the inherent connections between learning resources, lack the learner's personal learning style and knowledge level, violate the principle of "teaching according to aptitude", and ignore the adaptation process from learner to learning resources. Summary of the Invention

[0005] This application provides a method, apparatus, device, and storage medium for recommending learning resources based on knowledge graphs, in order to overcome the deficiencies of the aforementioned related technologies. The technical solution is as follows:

[0006] In a first aspect, embodiments of this application provide a learning resource recommendation method based on knowledge graphs, including the following steps:

[0007] Based on learners’ historical learning data, learner profiles and various learning resources used by the corresponding learners are obtained. Based on the learner profiles and various learning resources, a multimodal learning resource interaction map is generated to represent the set of learning resources used by any student.

[0008] Based on the multimodal learning resource interaction graph, a contrastive loss function is calculated as a constraint on the embedding representation of learners and learning resources.

[0009] Based on the interaction between learners and learning resources, different subgraphs are generated, and the network model is constructed by inputting the relationships to obtain the matching coefficient between learners and learning resources.

[0010] Based on the constraints of the embedded representations of the learners and learning resources, and based on the values ​​of the matching coefficients, the learners' interest in various types of learning resources is calculated.

[0011] Based on the level of interest, various learning resources are ranked, and learning resources are recommended to the corresponding learners in turn based on the ranking results.

[0012] In one alternative embodiment of the first aspect, calculating the contrastive loss function based on the multimodal learning resource interaction graph includes:

[0013] The knowledge graph enhancement module performs adjacency matrix decomposition on any multimodal learning resource interaction graph, outputting a fully connected knowledge graph. The graph convolutional network feature learning module traverses and learns the original multimodal learning resource interaction graph and the corresponding fully connected knowledge graph, and the contrastive loss function is calculated by comparison.

[0014] In one alternative embodiment of the first aspect, generating different subgraphs based on the interaction relationship between learners and learning resources, constructing a network model from the input relationships, and obtaining the matching coefficients between learners and learning resources includes:

[0015] A first relationship between learners and learning resources is determined, a second relationship between learners and learning resources is randomly generated, the similarity between the first relationship and the second relationship is calculated, the original multimodal learning resource interaction graph corresponding to the first relationship is updated based on the similarity, and the updated multimodal learning resource interaction graph is divided into three relationship subgraphs based on the learning resource classification module.

[0016] In one alternative embodiment of the first aspect, generating different subgraphs based on the interaction relationship between learners and learning resources, constructing a network model from the input relationships, and obtaining the matching coefficients between learners and learning resources includes:

[0017] The three relational subgraphs are input into the feature representation learning module, and the learning resource representation and learner representation are iteratively calculated based on the local weighted smoothing method and weighted propagation.

[0018] The learning resource representation and the learner representation are input into the high-order feature fusion module to obtain the representation vector of each layer of the learner and the learning resource, and the final representation vector is obtained by summing the representation vectors of each layer.

[0019] The matching coefficient between learners and learning resources is calculated based on the inner product of the learner's final representation vector and the learning resource's final representation vector.

[0020] Secondly, embodiments of this application also provide a knowledge graph-based learning resource recommendation device, comprising:

[0021] The data acquisition module is used to acquire learner profiles and various learning resources used by learners based on their historical learning data, and to generate a multimodal learning resource interaction map based on the learner profiles and various learning resources, which is used to represent the set of learning resources used by any student.

[0022] The loss function module is used to calculate a contrastive loss function based on the multimodal learning resource interaction graph, which serves as a constraint on the embedding representation of learners and learning resources.

[0023] Based on the interaction between learners and learning resources, different subgraphs are generated, and the network model is constructed by inputting the relationships to obtain the matching coefficient between learners and learning resources.

[0024] The relation calculation module is used to calculate the learner's interest in various learning resources based on the constraints of the embedded representations of the learner and the learning resources, and based on the value of the matching coefficient.

[0025] The resource recommendation module is used to sort various learning resources based on the degree of interest, and recommend learning resources to the corresponding learners in turn based on the sorting results.

[0026] In one alternative of the second aspect, the loss function module is further configured to perform adjacency matrix decomposition on any multimodal learning resource interaction graph based on the knowledge graph enhancement module, output a fully connected knowledge graph, and perform traversal learning on the original multimodal learning resource interaction graph and the corresponding fully connected knowledge graph based on the graph convolutional network learning feature module, and calculate the contrastive loss function by comparison.

[0027] Thirdly, embodiments of this application also provide an electronic device, the electronic device including a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the method provided by the first aspect or any implementation of the first aspect of the embodiments of this application.

[0028] Fourthly, this application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided by the first aspect of the embodiments of this application or any implementation thereof.

[0029] The beneficial effects of the technical solutions provided in some embodiments of this application include at least the following:

[0030] (1) It takes into account both the multimodal information fusion of learning resources and the learner's historical learning data, and integrates the two into the learning resource interaction network, which enriches its expressive meaning, improves the accuracy of learning resource adaptation, and provides a customized intelligent learning guidance solution.

[0031] (2) Provide knowledge graph enhancement methods to explore the connections of fully connected knowledge graphs, and use contrastive learning loss function to constrain the embedded representations of learners and learning resources, so as to make the recommended learning resources more accurate.

[0032] (3) Three virtual relationships were constructed to classify the connection between learners and learning resources and form subgraphs. By learning different aspects of the subgraphs, the intrinsic connection of learning resources was explored, which improved the accuracy, stability and effectiveness of learning resource adaptation.

[0033] (4) The attention mechanism is used to combine first-order features from different subgraphs and fuse higher-order features, making full use of the information of the learning resource interaction network, which helps to learn the final embedded representation of the entity. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart illustrating a knowledge graph-based learning resource recommendation method according to an embodiment of this application;

[0036] Figure 2 This is a flowchart illustrating a knowledge graph-based learning resource recommendation method according to an embodiment of this application.

[0037] Figure 3 This is a flowchart illustrating a knowledge graph-based learning resource recommendation method provided in an embodiment of this application.

[0038] Figure 4 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0040] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or modules is not limited to the steps or modules listed, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to such process, method, product, or apparatus.

[0041] It should be noted that the terms "first" and "second" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects. It is understood that "first" and "second" can be interchanged in a specific order or sequence where permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in an order other than those described or illustrated herein.

[0042] The present application will now be described in detail with reference to specific embodiments.

[0043] Next, combine Figure 1 This paper introduces a knowledge graph-based learning resource recommendation method provided in embodiments of this application. For details, please refer to... Figure 1 , Figure 1 This illustration shows a flowchart of a knowledge graph-based learning resource recommendation method provided in an embodiment of this application. The method includes the following steps:

[0044] S1 obtains the learner profile and the various learning resources used by the corresponding learner based on the learner's historical learning data, and generates a multimodal learning resource interaction map based on the learner profile and the various learning resources to represent the set of learning resources used by any student.

[0045] Specifically, learner profiles with obvious label parameters can be collected, along with text, images, and other learning resources. Weighted label features of different learning styles are used as the learner's initial embedding representation, and multimodal information of learning resources is aggregated to form a multimodal learning resource interaction network based on the learner's historical learning situation.

[0046] Understandably, student learning profiles can be inferred based on historical learning data, and student learning habits can be analyzed based on long-term learning statistics, thereby inferring learner profiles.

[0047] S2 calculates a contrastive loss function based on the multimodal learning resource interaction graph, which serves as a constraint on the embedding representation of learners and learning resources.

[0048] like Figure 2 As shown, according to the above scheme, the implementation steps of the knowledge graph enhancement module are as follows:

[0049] Using a multimodal learning resource interaction graph as input, for each learner's u i and learning resources i Assign an embedding vector Where d is the embedding size. The set of all learner and learning resource embeddings is defined as E. (u) ∈R I×d and E (s) ∈R J×d , where I and J represent the number of learners and learning resources, respectively.

[0050] Optionally, a stochastic SVD algorithm can be used, which first approximates the range of the input matrix with a low-rank orthogonal matrix, and then performs SVD decomposition on this smaller matrix.

[0051] Specifically, A is the normalized adjacency matrix. After performing singular value decomposition on A, we obtain three matrices of rank r: U r S r V r =RandSVD(A,r), where r is the rank required to decompose the matrix. RandSVD is a set of singular value decomposition methods built into PyTorch.

[0052] Furthermore, in order to "restore" matrix A, the decomposed matrix can be obtained by applying the following formula:

[0053]

[0054] The new adjacency matrix offers two advantages over the old one: firstly, it emphasizes the principal components of the knowledge graph by identifying learner preference representations. Secondly, the generated new graph structure preserves global collaboration signals by considering each learner-learning resource interaction pair.

[0055] Furthermore, the input graph convolutional network learns the feature module:

[0056] On the one hand, it is necessary to perform graph convolutional traversal learning on the original learning resource interaction network.

[0057] The specific process is as follows:

[0058] Perform a random edge-dropping operation on the adjacency matrix A, denoted as p. The adjacency matrix after random edge-dropping is represented as p(A). i ,:) and p(:,A j ).

[0059] From the learner's perspective, each learner node needs to aggregate neighboring learning resource nodes, which can be represented as follows: Similarly, from the perspective of related learning resource nodes, this aggregation process can be represented as follows: l indicates that the aggregation process occurs in the l-th layer of the graph convolutional neural network.

[0060] Furthermore, to ensure that the differences in preference levels are not too large, the above aggregation representation needs to be corrected, i.e., by nesting an activation function, which can be further expressed by the following formula:

[0061]

[0062] The activation function is denoted by σ(·), specifically the LeakyReLu function with a negative slope of 0.5. Let this be the representation of learner node u at layer l. This represents the layer l representation of the associated learning resource node s.

[0063] On the other hand, graph convolutional traversal learning is needed on the enhanced knowledge graph guided by singular value decomposition to obtain the approximate matrix and the collective representation of the embeddings, as follows:

[0064]

[0065] in These are the sets of embeddings of learner nodes and associated learning resource nodes encoded from the newly generated graph structure view.

[0066] According to the above scheme, the computation graph contrastive loss function module takes the learning resource interaction network and the fully connected interaction network as inputs, selects the InfoNCE loss function to optimize the contrastive learning process, and then obtains the contrastive loss function.

[0067] S3 generates different subgraphs based on the interaction between learners and learning resources, inputs the relationships to build a network model, and obtains the matching coefficients between learners and learning resources.

[0068] Specifically, a first relationship between learners and learning resources is determined, a second relationship between learners and learning resources is randomly generated, the similarity between the first relationship and the second relationship is calculated, the original multimodal learning resource interaction graph corresponding to the first relationship is updated based on the similarity, and the updated multimodal learning resource interaction graph is divided into three relationship subgraphs based on the learning resource classification module.

[0069] Furthermore, the three relational subgraphs are input into the feature representation learning module, and the learning resource representation and learner representation are iteratively calculated based on the local weighted smoothing method and weighted propagation.

[0070] The learning resource representation and the learner representation are input into the high-order feature fusion module to obtain the representation vector of each layer of the learner and the learning resource, and the final representation vector is obtained by summing the representation vectors of each layer.

[0071] The matching coefficient between learners and learning resources is calculated based on the inner product of the learner's final representation vector and the learning resource's final representation vector.

[0072] In a specific embodiment, such as Figure 3 As shown, the steps of the learning resource classification module include:

[0073] To cluster different relationships, i.e., different ways of acquiring learning resources, an unsupervised learning method is used to explore the latent factors of each original relationship, and original relationships with similar latent factors are merged into a virtual relationship. First, the representation of this virtual relationship is initialized as a virtual centroid matrix V∈R. K×d :

[0074] V = (v1, v2, ..., v K ) T ;

[0075] Where the k-th row v k ∈R d It is the kth virtual relation, where K is the number of virtual relations and is a hyperparameter that is set to 3 by default.

[0076] The original relationship is the first relationship between the learner and the learning resource. The virtual relationship is the second relationship between the learner and the learning resource, which can be obtained by replacing some values ​​in the first relationship through mapping.

[0077] Furthermore, the similarity between original and virtual relations in the knowledge graph is calculated. The correlation between each learner-learning resource interaction pair and the components of the obtained centroid matrix is ​​calculated for relation r. p ∈R d Its similarity vector s was constructed. p ∈R d Apply the formula:

[0078] s p =(m(r p ,v1),m(r p ,v2),…,m(r p ,v K ));

[0079] m(r p ,v k ) = r p T v k ;

[0080] The learning resource interaction network is divided into three subgraphs to capture learning resources acquired through similar methods.

[0081] The relation r is expressed by the following formula. p Replaced with the most similar virtual relationship v k′ Implement knowledge graph triples (h, r) p ,t) to (h,v k′ The conversion of ,t).

[0082] k′=argmax s p =argmax(m(r p ,v1),m(r p ,v2),…,m(r p ,v K ));

[0083] Where h is the head entity and t is the tail entity.

[0084] Furthermore, the steps of the feature representation learning module are as follows:

[0085] First, a local weighted smoothing method is used. For each sample point to be predicted, different weights are assigned to the sample points around that point according to the similarity between the neighboring nodes. For each node, its local embedding representation is learned. express:

[0086]

[0087] in It is the ID embedding of the learning resource entity as the tail entity, t,π(h,t) are the weights in the local information propagation, and N k (h) represents all head entities in the k-th subgraph, and the weights reflect the contribution of entity t to describing entity h.

[0088]

[0089] Secondly, the aggregate entity represents e. h and local information representation To obtain a temporary representation of the entity as its final representation.

[0090]

[0091] Further, normalization is performed using the following formula:

[0092]

[0093] Using fagg(·) to represent the aggregation function, each node of each subgraph is represented as:

[0094]

[0095] To further modify the node representation so that the distance in the embedding space better reflects similarity, a temporary representation is used. Iteratively smoothing the node vector h involves repeating the weighted propagation steps described above:

[0096]

[0097] Where p = 2, 3…Q, and Q = 3 are default values. This updates the entity representation as follows:

[0098]

[0099] The steps of the smoothing operation that incorporates the first-order information into the node embedding are represented as f. LWS (·):

[0100]

[0101] Furthermore, to maintain the semantic independence of item attributes, each subgraph is encoded with learning resource nodes from K different aspects, and then these are fused together using a K-attention mechanism to obtain... The final learning resource representation:

[0102]

[0103] To incorporate information from higher-order neighbors, multiple LWS aggregation layers are stacked. After l-layer aggregation, the entity representation is as follows:

[0104]

[0105] Similarly, the representation of the learner node can be obtained directly:

[0106]

[0107] It includes its l-hop neighbors, and l = 2, 3...L, where L defaults to 3.

[0108] Furthermore, the steps of the higher-order feature fusion module are as follows:

[0109] Obtain the representation vectors for each layer of the learner and learning resources. Finally, aggregate the outputs of each layer into the final representation.

[0110]

[0111] The inner product of the embedding vectors of the learner and the learning resources is obtained. The matching coefficient between learners and learning resources is expressed by the formula:

[0112]

[0113] Specifically, the result of the above formula is a value between 0 and 1.

[0114] S4 calculates the learner's level of interest in various learning resources based on the constraints of the embedded representations of the learner and learning resources, and based on the value of the matching coefficient.

[0115] Optionally, the data based on the matching coefficient is taken from the corresponding interest degree mapping table, and the interest degree on the corresponding learning resource is obtained according to the mapping relationship;

[0116] Optionally, different mapping tables can be set for different learning resources.

[0117] S5 sorts the various learning resources based on the degree of interest, and recommends learning resources to the corresponding learners in turn based on the sorting results.

[0118] Understandably, learning resources can be ranked according to the degree of interest, and when recommending learning resources to learners, those with higher interest levels can be given priority.

[0119] The recommendation of learning resources can include methods such as links, floating windows, and webpage layout. Links are sorted based on the sorting order of learning resources, the position and display order of floating windows are adjusted, the position of different learning resources in the webpage layout is adjusted, and the webpage is laid out based on the sorting order of learning resources.

[0120] Furthermore, it can collect learners' learning data in real time, monitor changes in learners' learning interests in different periods, calculate the degree of interest in different periods in real time, and adjust the ranking of learning resources.

[0121] The following are apparatus embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the method embodiments of this application.

[0122] This application also provides a knowledge graph-based learning resource recommendation device. This device can be implemented as a whole or part of a terminal through software, hardware, or a combination of both, or it can be integrated as an independent module on a server. The application usage time calculation device in this application embodiment can be applied to a terminal or the cloud. This device includes a data acquisition module, used to acquire a learner profile and various learning resources used by the learner based on the learner's historical learning data, and to generate a multimodal learning resource interaction graph based on the learner profile and the various learning resources, used to represent the set of learning resources used by any student.

[0123] The loss function module is used to calculate a contrastive loss function based on the multimodal learning resource interaction graph, which serves as a constraint on the embedding representation of learners and learning resources.

[0124] Based on the interaction between learners and learning resources, different subgraphs are generated, and the network model is constructed by inputting the relationships to obtain the matching coefficient between learners and learning resources.

[0125] The relation calculation module is used to calculate the learner's interest in various learning resources based on the constraints of the embedded representations of the learner and the learning resources, and based on the value of the matching coefficient.

[0126] The resource recommendation module is used to sort various learning resources based on the degree of interest, and recommend learning resources to the corresponding learners in turn based on the sorting results.

[0127] It should be noted that the knowledge graph-based learning resource recommendation device provided in the above embodiments is only illustrated by the division of the above functional modules when executing the knowledge graph-based learning resource recommendation method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the device provided in the above embodiments and the knowledge graph-based learning resource recommendation method embodiments belong to the same concept, and the implementation process is detailed in the method embodiments, which will not be repeated here.

[0128] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods described above.

[0129] Please see Figure 4 This is a structural block diagram of an electronic device provided in an embodiment of this application.

[0130] like Figure 4 As shown, the electronic device 400 includes a processor 401 and a memory 402.

[0131] In this embodiment, the processor 401 is the control center of the computer system, and can be a processor of a physical machine or a processor of a virtual machine. The processor 401 may include one or more processing cores, such as a 4-core processor or an 8-core processor. The processor 401 can be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array).

[0132] Processor 401 may also include a main processor and a coprocessor. The main processor is a processor used to process data in the wake-up state, also known as a CPU (Central Processing Unit). The coprocessor is a low-power processor used to process data in the standby state.

[0133] Memory 402 may include one or more computer-readable storage media, which may be non-transitory. Memory 402 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments of this application, the non-transitory computer-readable storage media in memory 402 is used to store at least one instruction, which is executed by processor 401 to implement the method in the embodiments of this application.

[0134] In some embodiments, the electronic device 400 further includes a peripheral device interface 403 and at least one peripheral device. The processor 401, memory 402, and peripheral device interface 403 can be connected via a bus or signal line. Various peripheral devices can be connected to peripheral device interface 403 via a bus, signal line, or circuit board. Specifically, the peripheral devices include: the display screen 404, camera 405, and audio circuitry 406. Peripheral device interface 403 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 401 and memory 402.

[0135] In some embodiments of this application, the processor 401, memory 402, and peripheral device interface 403 are integrated on the same chip or circuit board; in other embodiments of this application, any one or two of the processor 401, memory 402, and peripheral device interface 403 can be implemented on separate chips or circuit boards. This application does not specifically limit the implementation in this regard.

[0136] Display screen 404 is used to display a user interface (UI). This UI may include graphics, text, icons, videos, and any combination thereof. When display screen 404 is a touch display, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 401 for processing. In this case, display screen 404 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard.

[0137] In some embodiments of this application, there may be one display screen 404, disposed on the front panel of the electronic device 400; in other embodiments, there may be at least two display screens 404, disposed on different surfaces of the electronic device 400 or in a folded design; in still other embodiments, the display screen 404 may be a flexible display screen, disposed on a curved or folded surface of the electronic device 400. Furthermore, the display screen 404 may be configured as a non-rectangular, irregular shape, i.e., a non-rectangular screen. The display screen 404 may be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0138] Camera 405 is used to capture images or videos. Optionally, camera 405 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of the electronic device, and the rear-facing camera is located on the back of the electronic device. In some embodiments, there are at least two rear-facing cameras, which are any one of a main camera, a depth-sensing camera, a wide-angle camera, and a telephoto camera, to achieve background blurring by fusion of the main camera and the depth-sensing camera, panoramic shooting by fusion of the main camera and the wide-angle camera, VR (Virtual Reality) shooting, or other fusion shooting functions. In some embodiments of this application, camera 405 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cool light flash, which can be used for light compensation at different color temperatures.

[0139] The audio circuit 406 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals that are input to the processor 401 for processing. For stereo sound acquisition or reduction purposes, there may be multiple microphones, which are respectively located in different parts of the electronic device 400. The microphone may also be an array microphone or an omnidirectional microphone.

[0140] Power supply 407 is used to supply power to various components in electronic device 400. Power supply 407 can be alternating current, direct current, a disposable battery, or a rechargeable battery. When power supply 407 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged via a wired line, while a wireless rechargeable battery is a battery that is charged via a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0141] The block diagram of the electronic device shown in the embodiments of this application does not constitute a limitation on the electronic device 400. The electronic device 400 may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0142] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the methods in any of the foregoing embodiments. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0143] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A learning resource recommendation method based on knowledge graphs, characterized in that, include: Based on learners’ historical learning data, learner profiles and various learning resources used by the corresponding learners are obtained. Based on the learner profiles and various learning resources, a multimodal learning resource interaction map is generated to represent the set of learning resources used by any student. Based on the multimodal learning resource interaction graph, a contrastive loss function is calculated as a constraint on the embedding representation of learners and learning resources. Based on the interaction between learners and learning resources, different subgraphs are generated. These subgraphs are then used to construct a network model, and the matching coefficients between learners and learning resources are obtained, including: A first relationship between learners and learning resources is determined, a second relationship between learners and learning resources is randomly generated, the similarity between the first relationship and the second relationship is calculated, the original multimodal learning resource interaction graph corresponding to the first relationship is updated based on the similarity, and the updated multimodal learning resource interaction graph is divided into three relationship subgraphs based on the learning resource classification module. The three relational subgraphs are input into the feature representation learning module, and the learning resource representation and learner representation are iteratively calculated based on the local weighted smoothing method and weighted propagation. The learning resource representation and the learner representation are input into the high-order feature fusion module to obtain the representation vector of each layer of the learner and the learning resource, and the final representation vector is obtained by summing the representation vectors of each layer. The matching coefficient between learners and learning resources is calculated based on the inner product of the learner's final representation vector and the learning resource's final representation vector. Based on the constraints of the embedded representations of the learners and learning resources, and based on the values ​​of the matching coefficients, the learners' interest in various types of learning resources is calculated. Based on the level of interest, various learning resources are ranked, and learning resources are recommended to the corresponding learners in turn based on the ranking results.

2. The method according to claim 1, characterized in that, The calculation of the contrastive loss function based on the multimodal learning resource interaction graph includes: The knowledge graph enhancement module performs adjacency matrix decomposition on any multimodal learning resource interaction graph, outputting a fully connected knowledge graph. The graph convolutional network feature learning module traverses and learns the original multimodal learning resource interaction graph and the corresponding fully connected knowledge graph, and the contrastive loss function is calculated by comparison.

3. An apparatus for recommending learning resources based on knowledge graphs as described in any one of claims 1-2, characterized in that, include: The data acquisition module is used to acquire learner profiles and various learning resources used by learners based on their historical learning data, and to generate a multimodal learning resource interaction map based on the learner profiles and various learning resources, which is used to represent the set of learning resources used by any student. The loss function module is used to calculate a contrastive loss function based on the multimodal learning resource interaction graph, which serves as a constraint on the embedding representation of learners and learning resources. Based on the interaction between learners and learning resources, different subgraphs are generated, and the network model is constructed by inputting the relationships to obtain the matching coefficient between learners and learning resources. The relation calculation module is used to calculate the learner's interest in various learning resources based on the constraints of the embedded representations of the learner and the learning resources, and based on the value of the matching coefficient. The resource recommendation module is used to sort various learning resources based on the degree of interest, and recommend learning resources to the corresponding learners in turn based on the sorting results.

4. The apparatus according to claim 3, characterized in that, include: The loss function module is also used to perform adjacency matrix decomposition on any multimodal learning resource interaction graph based on the knowledge graph enhancement module, output a fully connected knowledge graph, and perform traversal learning on the original multimodal learning resource interaction graph and the corresponding fully connected knowledge graph based on the graph convolutional network learning feature module, and calculate the contrastive loss function by comparison.

5. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 2.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 2.

Images