Knowledge recommendation method and device of online education platform, computer device and medium

Abstract

The application discloses a kind of knowledge recommendation method, device, computer equipment and medium of online education platform, comprising: constructing the heterogeneous information network of online education platform, and dynamically generating the semantic embedding of knowledge concept, obtain basic embedding information;The enhancement optimization of knowledge concept is carried out to basic embedding information, and enhanced embedding information is obtained;Based on enhanced embedding information, construct the multi-type user interaction graph that integrates time context information;Message transmission is carried out in multi-type user interaction graph, and the node embedding of user and knowledge concept is updated, and target embedding information is obtained;Different sources of target embedding information are weighted and fused by semantic attention module, and fusion embedding information is obtained;The preference of user to knowledge concept is predicted and recommendation information is generated using extended matrix decomposition model and fusion embedding information, and the accuracy of knowledge recommendation of online education platform is improved using the application.

Description

Technical Field

[0001] The present invention relates to the field of information recommendation technology, and in particular to a knowledge recommendation method, device, computer equipment and medium for an online education platform. Background Art

[0002] With the development of artificial intelligence technology, online education platforms are becoming increasingly widespread. Their advantages, such as abundant resources and freedom from time and space constraints, have attracted a growing number of students and teachers. However, as the number of resources continues to increase, it becomes increasingly difficult for learners to find the content they truly interest or need. Research shows that many learners lack the perseverance and motivation to complete their courses, resulting in the majority failing to complete their studies and causing high dropout rates.

[0003] The current approach is to create a user profile and then determine user preferences to make recommendations. While developing this invention, the inventors discovered that while this existing approach can provide targeted recommendations to a certain extent, it can lead to inaccurate profiles and recommendations when data is scarce. Furthermore, the interests of the same user can change over time, indicating that this approach lacks high accuracy in knowledge recommendation. Summary of the Invention

[0004] Embodiments of the present invention provide a knowledge recommendation method, apparatus, computer equipment, and storage medium for an online education platform, so as to improve the accuracy of knowledge recommendation on the online education platform.

[0005] In order to solve the above technical problems, the present invention provides a knowledge recommendation method for an online education platform, including:

[0006] Constructing a heterogeneous information network of an online education platform, and using a meta-path-based random walk algorithm in the heterogeneous information network to dynamically generate semantic embeddings of knowledge concepts to obtain basic embedding information;

[0007] Using a neural network to enhance and optimize the knowledge concept of the basic embedded information to obtain enhanced embedded information;

[0008] Based on the enhanced embedding information, a multi-type user interaction graph incorporating time context information is constructed, wherein the multi-type user interaction graph records various interaction behaviors between users and knowledge concepts at different time points;

[0009] By using a graph neural network to pass messages in the multi-type user interaction graph, the node embeddings of users and knowledge concepts are updated, and the updated information of each enhanced embedding information is obtained as the target embedding information;

[0010] The target embedding information from different sources is weightedly fused through the semantic attention module to obtain the fused embedding information;

[0011] The extended matrix factorization model and the fused embedding information are used to predict user preferences for knowledge concepts and generate recommendation information.

[0012] Optionally, in the heterogeneous information network, a random walk algorithm based on a meta-path is used to dynamically generate semantic embeddings of knowledge concepts, and the basic embedding information obtained includes:

[0013] A random walk based on meta-path is used to sample heterogeneous information networks and filter out nodes of the same type. Each group of nodes of the same type is formed into a node sequence.

[0014] The node sequence is projected into a low-dimensional space using a skip-gram model to obtain a distributed representation of the node as the basic embedding information.

[0015] Optionally, the step of using a neural network to enhance and optimize the basic embedded information in terms of knowledge concepts to obtain enhanced embedded information includes:

[0016] Using a graph convolutional network to aggregate neighbor node information of the basic embedding information to generate an auxiliary embedding representation;

[0017] A graph attention network is used to aggregate neighbor node information on the basic embedding information to generate an enhanced embedding representation;

[0018] The auxiliary embedding representation, the enhanced embedding representation and the basic embedding information are spliced ​​and fused respectively to obtain first enhanced embedding information and second enhanced embedding information, and the first enhanced embedding information and the second enhanced embedding information are used as the enhanced embedding information.

[0019] Optionally, constructing a multi-type user interaction graph incorporating temporal context information based on the enhanced embedding information includes:

[0020] Construct the initial interaction graph G m =(V m ,E m ), if user u and item v interact under the kth interaction type, then u and v k There is an edge between them, so V m =U∪V',v k ∈V', U represents all users, V' is the item with interaction type considered, where the initial interaction graph G m There are (I+J×K) vertices in , where I and J represent the number of users and items respectively, and K represents the number of interaction types;

[0021] If user u has an item vk The interaction time is It is mapped to a single time slot The sine function is used to generate the relative time embedding of the interaction, and a multi-type user interaction graph that incorporates temporal context information is obtained.

[0022] Optionally, performing weighted fusion of target embedding information from different sources through a semantic attention module to obtain fused embedding information includes:

[0023] Calculate the attention weights of the target embedding information corresponding to user u and item v and perform normalization;

[0024] The normalized attention weights are applied to the target embedding information to obtain the fused embedding information.

[0025] Optionally, the knowledge recommendation method of the online education platform further includes:

[0026] Define the latent factors of user embedding and item embedding, and adjust the parameters of the initial matrix decomposition model through the optimization function;

[0027] The model is trained using a gradient descent method until convergence, thereby obtaining the extended matrix decomposition model.

[0028] In order to solve the above technical problems, the embodiment of the present application further provides a knowledge recommendation device for an online education platform, comprising:

[0029] An information extraction module is used to construct a heterogeneous information network of the online education platform and dynamically generate semantic embeddings of knowledge concepts in the heterogeneous information network using a meta-path-based random walk algorithm to obtain basic embedding information;

[0030] An information enhancement module, configured to enhance and optimize the basic embedded information in terms of knowledge concepts using a neural network to obtain enhanced embedded information;

[0031] An interaction graph construction module is used to construct a multi-type user interaction graph incorporating time context information based on the enhanced embedding information, wherein the multi-type user interaction graph records various interaction behaviors between users and knowledge concepts at different time points;

[0032] An information updating module is configured to perform message passing in the multi-type user interaction graph through a graph neural network, update the node embeddings of users and knowledge concepts, and obtain updated information of each enhanced embedding information as target embedding information;

[0033] The information fusion module is used to perform weighted fusion of target embedding information from different sources through the semantic attention module to obtain fused embedding information;

[0034] The knowledge recommendation module is used to predict the user's preference for knowledge concepts and generate recommendation information by using the extended matrix decomposition model and the fused embedding information.

[0035] Optionally, the information extraction module includes:

[0036] A node screening unit is used to sample the heterogeneous information network using a random walk based on meta-paths and screen out nodes of the same type, forming each group of nodes of the same type into a node sequence;

[0037] A node projection unit is used to project the node sequence into a low-dimensional space using a skip-gram model to obtain a distributed representation of the node as the basic embedding information.

[0038] Optionally, the information enhancement module includes:

[0039] A first aggregation unit is configured to aggregate neighbor node information of the basic embedding information using a graph convolutional network to generate an auxiliary embedding representation;

[0040] A second aggregation unit is configured to aggregate neighbor node information of the basic embedding information using a graph attention network to generate an enhanced embedding representation;

[0041] A fusion enhancement unit is used to splice and fuse the auxiliary embedding representation, the enhanced embedding representation and the basic embedding information respectively to obtain first enhanced embedding information and second enhanced embedding information, and use the first enhanced embedding information and the second enhanced embedding information as the enhanced embedding information.

[0042] In order to solve the above technical problems, an embodiment of the present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, the steps of the knowledge recommendation method of the above online education platform are implemented.

[0043] In order to solve the above technical problems, an embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the knowledge recommendation method of the above online education platform.

[0044] The knowledge recommendation method and device of the online education platform, the computer device and the storage medium provided by the embodiment of the application, by constructing a heterogeneous information network of the online education platform, and in the heterogeneous information network, using a random walk algorithm based on a meta path, dynamically generating semantic embedding of knowledge concepts, obtaining basic embedding information; using a neural network to enhance and optimize the knowledge concepts of the basic embedding information, obtaining enhanced embedding information; based on the enhanced embedding information, constructing a multi-type user interaction graph integrating time context information, the multi-type user interaction graph recording multiple interaction behaviors of users and knowledge concepts at different time points; through message passing in the multi-type user interaction graph by a graph neural network, updating the node embedding of the users and the knowledge concepts, obtaining the update information of each enhanced embedding information as target embedding information; through a semantic attention module, weighting and fusing the target embedding information of different sources, obtaining fused embedding information; using an extended matrix decomposition model and the fused embedding information, predicting the preference of the user for the knowledge concept and generating recommendation information, realizing semantic extraction according to the heterogeneous information network and the multi-type interaction data of the user, so as to obtain the user preference for accurate recommendation, and improving the knowledge recommendation accuracy of the online education platform. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the application. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0046] Figure 1 is an exemplary system architecture diagram to which the present application can be applied;

[0047] Figure 2 is a flowchart of one embodiment of the knowledge recommendation method of the online education platform of the present application;

[0048] Figure 3 is a structural example diagram of a meta path encoder according to the present application;

[0049] Figure 4 is a structural schematic diagram of one embodiment of the knowledge recommendation device of the online education platform according to the present application;

[0050] Figure 5 is a structural schematic diagram of one embodiment of the computer device according to the present application. DETAILED DESCRIPTION

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of the application are for the purpose of describing specific embodiments only and are not intended to limit this application. The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0052] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0054] See also Figure 1 ,like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105. Network 104 is a medium for providing communication links between terminal devices 101, 102, 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0055] Users can use terminal devices 101 , 102 , 103 to interact with server 105 via network 104 to receive or send messages, etc.

[0056] Terminal devices 101, 102, and 103 can be various electronic devices with display screens and support web browsing, including but not limited to smartphones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV), laptop computers, desktop computers, etc.

[0057] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0058] It should be noted that the knowledge recommendation method of the online education platform provided in the embodiment of the present application is executed by the server, and accordingly, the knowledge recommendation device of the online education platform is set in the server.

[0059] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only illustrative. According to the implementation requirements, there can be any number of terminal devices, networks and servers. The terminal devices 101, 102, 103 in the embodiment of the present application can specifically correspond to application systems in actual production.

[0060] See also Figure 2 , Figure 2 The present invention provides a knowledge recommendation method for an online education platform. Figure 1 The server in the example is used as an example, and the details are as follows:

[0061] S201: Construct a heterogeneous information network for the online education platform, and use a meta-path-based random walk algorithm in the heterogeneous information network to dynamically generate semantic embeddings of knowledge concepts and obtain basic embedding information.

[0062] In a specific embodiment, in step S201, a random walk algorithm based on meta-path is used in a heterogeneous information network to dynamically generate semantic embeddings of knowledge concepts, and the basic embedding information obtained includes:

[0063] A random walk based on meta-path is used to sample heterogeneous information networks and filter out nodes of the same type. Each group of nodes of the same type is formed into a node sequence.

[0064] The skip-gram model is used to project the node sequence into a low-dimensional space to obtain the distributed representation of the node as the basic embedding information.

[0065] Specifically, if Figure 3 As shown, Figure 3 This is a schematic diagram of a specific meta-path encoder structure of this embodiment. This embodiment considers a heterogeneous information network G = {V, E}, uses a meta-path-based random walk to sample the heterogeneous information network and filter out node sequences of the same type. The walk strategy is shown in formula (1):

[0066]

[0067] In the above formula, represents the node visited in the i-th step of the random walk, t represents the type of the node, x i+1 Represents the node to be visited next. represent All neighbor nodes of type t+1, the probability calculation function P is used to calculate the probability of Walk to node x i+1 probability.

[0068] Furthermore, after obtaining the random walk path, the nodes of the same type are formed into a node sequence, and then the node sequence is projected into a low-dimensional space using the skip-gram model to learn the distributed representation of the nodes. Specifically, the optimization goal of the skip-gram model is Σ c∈C(x) logP(c t |θ(x)), where x represents the current node, X represents all nodes in the graph, and t represents the node

[0069] Type, T represents all node types, C(x) represents the node sequence obtained by random walk, and θ represents the mapping function. By applying this function to each entity in the sequence, the embedding h of the entity under the meta-path p can be obtained. p .

[0070] S202: Using a neural network to enhance and optimize the knowledge concepts of the basic embedded information to obtain enhanced embedded information.

[0071] In a specific embodiment, in step S202, a neural network is used to enhance and optimize the knowledge concept of the basic embedded information, and the enhanced embedded information obtained includes:

[0072] Use graph convolutional networks to aggregate neighbor node information of basic embedding information to generate auxiliary embedding representations;

[0073] A graph attention network is used to aggregate neighbor node information from the basic embedding information to generate an enhanced embedding representation;

[0074] The auxiliary embedding representation, the enhanced embedding representation and the basic embedding information are spliced ​​and fused respectively to obtain first enhanced embedding information and second enhanced embedding information, and the first enhanced embedding information and the second enhanced embedding information are used as enhanced embedding information.

[0075] Specifically, this embodiment uses a two-stage meta-path embedding enhancement module to optimize the embedding representation.

[0076] In the first stage, a single-layer GCN is used to aggregate the neighbors of the nodes to generate another set of auxiliary embeddings to enrich the number of meta-path embeddings. The adjacency matrix M is defined to describe the connection relationship between nodes in the context node set previously obtained by random walk. In order to consider self-connections, M is added with an identity matrix: M = M + I, and M is normalized using the degree matrix Δ: Finally, the generation function of auxiliary embedding is shown in formula (2):

[0077]

[0078] in and are two trainable weights, h p Represents the original meta-path embedding.

[0079] In the second stage, the graph attention network is used to further aggregate the node’s neighbor information to more comprehensively enhance and capture the node features. Taking item v as an example, the embedding of item v under meta-path l is The attention scores of its neighbor nodes are: η (l)T represents the attention parameter under the meta-path l, W (l) It is a trainable weight matrix. After calculating the attention scores of all neighbor nodes, it is normalized as shown in formula (3):

[0080]

[0081] After obtaining the normalized attention scores, a multi-head attention mechanism is used to calculate the new features of item v, as shown in formula (4):

[0082]

[0083] Among them, σ is a nonlinear function, [W p ] k is a trainable weight matrix under attention head k. After the calculation is completed, the results under all heads are spliced ​​together. In addition, in order to preserve the information of the original embedding, the original embedding and the embedding processed by GAT are spliced ​​together to obtain the final embedding: The auxiliary embedding obtained in the first stage is also processed in a similar way to obtain the enhanced

[0084] S203: Based on the enhanced embedding information, a multi-type user interaction graph incorporating time context information is constructed, where the multi-type user interaction graph records various interaction behaviors between users and knowledge concepts at different time points.

[0085] In this embodiment, a multi-type interaction graph incorporating time context information is constructed, and a graph neural network is used to perform message passing to update node embeddings, thereby obtaining a deep embedding representation that carries time series and interaction type information.

[0086] In a specific embodiment, in step S203, constructing a multi-type user interaction graph incorporating temporal context information based on the enhanced embedding information includes:

[0087] Construct the initial interaction graph G m =(V m ,E m ), if user u and item v interact under the kth interaction type, then u and v k There is an edge between them, so V m =U∪V',v k ∈V', U represents all users, V' is the item with interaction type considered, where the initial interaction graph G m There are (I+J×K) vertices in , where I and J represent the number of users and items respectively, and K represents the number of interaction types;

[0088] If user u has an item v k The interaction time is It is mapped to a single time slot The sine function is used to generate the relative time embedding of the interaction, and a multi-type user interaction graph that incorporates temporal context information is obtained.

[0089] Specifically, construct a multi-type interaction graph G m =(V m ,E m ). If user u and item v interact under the kth interaction type, then u and v k There is an edge between them, so V m =U∪V',v k ∈V', U represents all users, and V' is the item with the interaction type taken into account. Multi-type interaction graph G m There are (I+J×K) vertices in , where I and J represent the number of users and items respectively, and K represents the number of interaction types.

[0090] At the same time, a temporal context encoding scheme is introduced to enable the model to dynamically integrate the interaction between users and items into the message aggregation architecture. Specifically, suppose user u has an item v k The interaction time is Map it to a single time slot Then a sine function is used to generate the relative time embedding of the interaction. As shown in formula (5):

[0091]

[0092] In formula (5), (2i) and (2i+1) represent the even and odd position indices in the time embedding, respectively.

[0093] S204: Message passing is performed in a multi-type user interaction graph through a graph neural network to update the node embeddings of users and knowledge concepts, and updated information of each enhanced embedding information is obtained as the target embedding information.

[0094] Specifically, after constructing the multi-type interaction graph with time context information, the following paradigm is stipulated: k Messages are passed between them, as shown in formula (6):

[0095]

[0096] In this formula, φ(·) represents a feature of the input and h u Function to perform encoding. is an attenuation factor that normalizes the propagation effect according to the node degree and can be defined as: where N u and They are nodes u and v respectively k Therefore, from the object v k The message delivery paradigm to user u can be further defined as shown in Equation (7):

[0097]

[0098] Assign user u to item v k The relative time embedding of the interaction between user u and item v k In the message passing paradigm between , as shown in formula (8), we can get:

[0099]

[0100] Among them, l represents the neural network layer, Representative node u iTherefore, the message passing paradigm can be further defined as shown in formula (9):

[0101]

[0102] in is a trainable weight matrix, Represents element-by-element addition. Finally, the results of each layer of the neural network are concatenated to generate the final user u and item v k The multi-type interaction embedding of is shown in formula (10):

[0103]

[0104] Among them, the summary of item v is Generated by a gating mechanism to distinguish the importance of different types of interactions.

[0105] S205: Perform weighted fusion of target embedding information from different sources through the semantic attention module to obtain fused embedding information.

[0106] In this embodiment, after obtaining the meta-path embeddings (including original embeddings and auxiliary embeddings) and multi-type interaction embeddings of users and items, a semantic attention module is designed to calculate their weights and fuse them.

[0107] In a specific embodiment, in step S201, the target embedding information from different sources is weightedly fused by the semantic attention module to obtain the fused embedding information including:

[0108] Calculate the attention weights of the target embedding information corresponding to user u and item v and perform normalization;

[0109] The normalized attention weights are applied to the target embedding information to obtain the fused embedding information.

[0110] Specifically, consider the obtained embedding sets of user u and item v: Taking user u as an example, the attention weight of each embedding can be calculated as shown in the following formula (11):

[0111]

[0112] in and are two trainable weight matrices, is an attention bias, This is the attention weight, which reflects the importance of meta-path embedding and multi-type interaction embedding to the recommendation task. After the calculation is completed, it is normalized to obtain Finally, the final embedding of user u is calculated by formula (12):

[0113]

[0114] The final embedding e of item v v It is also calculated in a similar way.

[0115] S206: Utilize the extended matrix decomposition model and fused embedding information to predict user preferences for knowledge concepts and generate recommendation information.

[0116] In a specific embodiment, before step S206, the knowledge recommendation method of the online education platform further includes:

[0117] Define the latent factors of user embedding and item embedding, and adjust the parameters of the initial matrix decomposition model through the optimization function;

[0118] The gradient descent method is used to train the model until convergence, and the extended matrix factorization model is obtained.

[0119] Specifically, an extended matrix decomposition module is introduced to generate prediction results and optimize the function, which can be defined as: where e u and e v represent user embedding and item embedding respectively, x u and y v Represent the latent factors of users and items respectively, γ v and γ u are two trainable parameters used to ensure that the user embedding e u and item embedding v In the same space, α and β are two adjustable parameters. The final optimization function is defined as shown in formula (13):

[0120]

[0121] To prevent overfitting, L2 regularization is added to the optimization function. Θ represents all trainable parameters in the model. Gradient descent is used during training, calculating partial derivatives and iteratively updating the parameters in the direction of the negative gradient until the model converges.

[0122] In this embodiment, a heterogeneous information network of an online education platform is constructed, and in the heterogeneous information network, a semantic embedding of a knowledge concept is dynamically generated using a random walk algorithm based on a meta path to obtain basic embedding information; a neural network is used to enhance and optimize the basic embedding information for the knowledge concept to obtain enhanced embedding information; based on the enhanced embedding information, a multi-type user interaction graph integrating time context information is constructed, and the multi-type user interaction graph records various interaction behaviors of users and knowledge concepts at different time points; a message is passed in the multi-type user interaction graph through a graph neural network to update node embedding of the users and the knowledge concepts to obtain updated information of each enhanced embedding information as target embedding information; different sources of target embedding information are weighted and fused through a semantic attention module to obtain fused embedding information; an extended matrix decomposition model and the fused embedding information are used to predict user preferences for knowledge concepts and generate recommendation information, so as to realize semantic extraction based on the heterogeneous information network and multi-type interaction data of the users, thereby obtaining user preferences for accurate recommendation and improving the knowledge recommendation accuracy of the online education platform.

[0123] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0124] Figure 4 The principle block diagram of the knowledge recommendation device of the online education platform corresponding to the knowledge recommendation method of the online education platform in the above embodiment is shown. As shown in Figure 4 The knowledge recommendation device of the online education platform includes an information extraction module 31, an information enhancement module 32, an interaction graph construction module 33, an information update module 34, an information fusion module 35, and a knowledge recommendation module 36. The functions of each module are described in detail as follows:

[0125] The information extraction module 31 is used to construct a heterogeneous information network of an online education platform, and in the heterogeneous information network, a semantic embedding of a knowledge concept is dynamically generated using a random walk algorithm based on a meta path to obtain basic embedding information;

[0126] The information enhancement module 32 is used to enhance and optimize the basic embedding information for the knowledge concept using a neural network to obtain enhanced embedding information;

[0127] The interaction graph construction module 33 is used to construct a multi-type user interaction graph integrating time context information based on the enhanced embedding information, and the multi-type user interaction graph records various interaction behaviors of users and knowledge concepts at different time points;

[0128] An information updating module 34 is configured to perform message passing in a multi-type user interaction graph through a graph neural network, update the node embeddings of users and knowledge concepts, and obtain updated information of each enhanced embedding information as target embedding information;

[0129] An information fusion module 35 is used to perform weighted fusion of target embedding information from different sources through a semantic attention module to obtain fused embedding information;

[0130] The knowledge recommendation module 36 is used to predict the user's preference for knowledge concepts and generate recommendation information by using the extended matrix decomposition model and fusion embedding information.

[0131] Optionally, the information extraction module 31 includes:

[0132] A node screening unit is used to sample the heterogeneous information network using a random walk based on meta-paths and screen out nodes of the same type, forming each group of nodes of the same type into a node sequence;

[0133] The node projection unit is used to project the node sequence into a low-dimensional space using a skip-gram model to obtain a distributed representation of the node as basic embedding information.

[0134] Optionally, the information enhancement module 32 includes:

[0135] The first aggregation unit is used to aggregate neighbor node information of the basic embedding information using a graph convolutional network to generate an auxiliary embedding representation;

[0136] The second aggregation unit is used to aggregate neighbor node information of the basic embedding information using the graph attention network to generate an enhanced embedding representation;

[0137] The fusion enhancement unit is used to splice and fuse the auxiliary embedding representation, the enhanced embedding representation and the basic embedding information respectively to obtain first enhanced embedding information and second enhanced embedding information, and use the first enhanced embedding information and the second enhanced embedding information as enhanced embedding information.

[0138] For the specific definition of the knowledge recommendation device of the online education platform, please refer to the definition of the knowledge recommendation method of the online education platform above, which will not be repeated here. The various modules in the knowledge recommendation device of the above-mentioned online education platform can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0139] To solve the above technical problems, the present application also provides a computer device. Figure 5 , Figure 5This is a basic structural block diagram of the computer device in this embodiment.

[0140] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components connected to the memory 41, the processor 42, and the network interface 43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0141] The computer device may be a desktop computer, notebook computer, PDA, cloud server, etc. The computer device may interact with the user via a keyboard, mouse, remote control, touchpad, or voice control device.

[0142] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (for example, SD or D interface display memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as the program code of the knowledge recommendation method of the online education platform. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0143] The processor 42 may, in some embodiments, be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 42 is generally used to control the overall operation of the computer device 4. In the present embodiment, the processor 42 is configured to run program codes or process data stored in the memory 41, such as program codes of the knowledge recommendation method of the online education platform.

[0144] The network interface 43 may include a wireless network interface or a wired network interface, and is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0145] The present application also provides another embodiment, i.e., a computer readable storage medium storing an interface display program, the interface display program being executable by at least one processor to cause the at least one processor to perform the steps of the knowledge recommendation method of the online education platform as described above.

[0146] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software and the necessary general hardware platform, of course, they can also be implemented by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a plurality of instructions for causing a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present application.

[0147] Obviously, the above-described embodiments are only some of the embodiments of the present application, not all the embodiments, and the preferred embodiments of the present application are given in the drawings, but do not limit the patent scope of the present application. The present application can be implemented in many different forms, and conversely, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing specific embodiments, or make equivalent replacements to some technical features. Any equivalent structure made by using the contents of the present application specification and drawings, directly or indirectly applied to other related technical fields, is also within the scope of the patent protection of the present application.

Claims

1. A knowledge recommendation method for an online education platform, characterized in that: include: Constructing a heterogeneous information network of an online education platform, and using a meta-path-based random walk algorithm in the heterogeneous information network to dynamically generate semantic embeddings of knowledge concepts to obtain basic embedding information; Using a neural network to enhance and optimize the knowledge concept of the basic embedded information to obtain enhanced embedded information; Based on the enhanced embedding information, a multi-type user interaction graph incorporating time context information is constructed, wherein the multi-type user interaction graph records various interaction behaviors between users and knowledge concepts at different time points; By using a graph neural network to pass messages on the multi-type user interaction graph, the node embeddings of users and knowledge concepts are updated, and the updated information of each enhanced embedding information is obtained as the target embedding information. After constructing the multi-type interaction graph with temporal context information, the message passing paradigm is defined as shown in formula (9): in and are two trainable weight matrices, Represents element-by-element addition, N u and They are node u and node v respectively k The degree, and Represents node u and node v respectively k The representation at neural network layer l is, is user u's response to item v k At time Finally, the results of each layer of the neural network are concatenated to generate the final user u and item v k The multi-type interaction embedding of is shown in formula (10): Among them, || represents splicing, Represents the multi-type interaction embedding of user u, Represents the embedding of the item under interaction type k, and the final summary representation of item v It is further generated by a gating mechanism; the target embedding information from different sources is weightedly fused through the semantic attention module to obtain the fused embedding information; The extended matrix factorization model and the fused embedding information are used to predict user preferences for knowledge concepts and generate recommendation information.

2. The knowledge recommendation method of the online education platform according to claim 1, characterized in that: In the heterogeneous information network, a random walk algorithm based on meta-path is used to dynamically generate semantic embeddings of knowledge concepts, and the basic embedding information obtained includes: A random walk based on meta-path is used to sample heterogeneous information networks and filter out nodes of the same type. Each group of nodes of the same type is formed into a node sequence. The node sequence is projected into a low-dimensional space using a skip-gram model to obtain a distributed representation of the node as the basic embedding information.

3. The knowledge recommendation method of the online education platform according to claim 1, characterized in that: The method of using a neural network to enhance and optimize the knowledge concept of the basic embedded information to obtain enhanced embedded information includes: Using a graph convolutional network to aggregate neighbor node information of the basic embedding information to generate an auxiliary embedding representation; A graph attention network is used to aggregate neighbor node information on the basic embedding information to generate an enhanced embedding representation; The auxiliary embedding representation, the enhanced embedding representation and the basic embedding information are spliced ​​and fused respectively to obtain first enhanced embedding information and second enhanced embedding information, and the first enhanced embedding information and the second enhanced embedding information are used as the enhanced embedding information.

4. The knowledge recommendation method of the online education platform according to claim 1, characterized in that: The step of constructing a multi-type user interaction graph incorporating temporal context information based on the enhanced embedding information includes: Construct the initial interaction graph G m =(V m ,E m ), if user u and item v interact under the kth interaction type, then u and v k There is an edge between them, so V m =U∪V',v k ∈V', U represents all users, V' is the item with interaction type considered, where the initial interaction graph G m There are (I+J×K) vertices in , where I and J represent the number of users and items respectively, and K represents the number of interaction types; If user u has an item v k The interaction time is It is mapped to a single time slot The sine function is used to generate the relative time embedding of the interaction, and a multi-type user interaction graph that incorporates temporal context information is obtained.

5. The knowledge recommendation method of the online education platform according to claim 1, characterized in that: The weighted fusion of target embedding information from different sources by the semantic attention module to obtain fused embedding information includes: Calculate the attention weights of the target embedding information corresponding to user u and item v and perform normalization; The normalized attention weights are applied to the target embedding information to obtain the fused embedding information.

6. The knowledge recommendation method of the online education platform according to claim 1, characterized in that: The knowledge recommendation method of the online education platform further includes: Define the latent factors of user embedding and item embedding, and adjust the parameters of the initial matrix decomposition model through the optimization function; The model is trained using a gradient descent method until convergence, thereby obtaining the extended matrix decomposition model.

7. A knowledge recommendation device for an online education platform, characterized in that: include: An information extraction module is used to construct a heterogeneous information network of the online education platform and dynamically generate semantic embeddings of knowledge concepts in the heterogeneous information network using a meta-path-based random walk algorithm to obtain basic embedding information; An information enhancement module, configured to enhance and optimize the basic embedded information in terms of knowledge concepts using a neural network to obtain enhanced embedded information; An interaction graph construction module is used to construct a multi-type user interaction graph incorporating time context information based on the enhanced embedding information, wherein the multi-type user interaction graph records various interaction behaviors between users and knowledge concepts at different time points; The information update module is used to perform message passing in the multi-type user interaction graph through the graph neural network, update the node embedding of users and knowledge concepts, and obtain the updated information of each enhanced embedding information as the target embedding information; wherein, after constructing the multi-type interaction graph with time context information, the message passing paradigm is defined as shown in formula (9): in, and are two trainable weight matrices, Represents element-by-element addition, N u and They are node u and node v respectively k The degree, and Represents node u and node v respectively k The representation at neural network layer l is, is user u's relationship with item v k At time Finally, the results of each layer of the neural network are concatenated to generate the final user u and item v k The multi-type interaction embedding of is shown in formula (10): Among them, || represents splicing, Represents the multi-type interaction embedding of user u, Represents the embedding of the item under interaction type k, and the final summary representation of item v further generated by a gating mechanism; The information fusion module is used to perform weighted fusion of target embedding information from different sources through the semantic attention module to obtain fused embedding information; The knowledge recommendation module is used to predict the user's preference for knowledge concepts and generate recommendation information by using the extended matrix decomposition model and the fused embedding information.

8. The knowledge recommendation device for an online education platform according to claim 7, characterized in that: The information extraction module includes: A node screening unit is used to sample the heterogeneous information network using a random walk based on meta-paths and screen out nodes of the same type, forming each group of nodes of the same type into a node sequence; A node projection unit is used to project the node sequence into a low-dimensional space using a skip-gram model to obtain a distributed representation of the node as the basic embedding information.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the knowledge recommendation method for the online education platform as described in any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the knowledge recommendation method for an online education platform as described in any one of claims 1 to 6 is implemented.

Images