Knowledge graph-based software-defined science and education resource dynamic collaborative recommendation method and system

Abstract

The method and system for dynamically recommending software-defined science and education resources based on a knowledge graph comprise the following steps: constructing an artificial intelligence knowledge graph; obtaining science and education resource data of a science and education institution, extracting knowledge entities in the science and education resource data, connecting the knowledge entities to the artificial intelligence knowledge graph, and obtaining a science and education resource knowledge graph; constructing a user preference network model containing graph updating and time sequence feature fusion, extracting local resource features and user features, and obtaining resource features and user features of each local client; constructing a local knowledge graph based on the science and education resource knowledge graph, using the resource features of each local client as initial features for model training until the model converges; inputting the user and the resource into a graph neural network for prediction, and recommending resources with high scores to each user. The application greatly widens the data source and coverage range of the recommendation system, and realizes truly multi-source science and education resource recommendation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of resource recommendation, and particularly relates to a software-defined science and education resource dynamic collaborative recommendation method and system based on a knowledge graph. BACKGROUND

[0002] A large amount of science and education resources are generated in the process of digital education, which greatly improves the resource acquisition efficiency of learners, but there are problems of resource complexity and difficulty in management. In view of this, a software-defined science and education resource management model is proposed, the skill points and knowledge points contained in the science and education resources are mined through a knowledge graph, and a science and education resource knowledge semantic network is constructed by taking the knowledge graph as a medium, so as to overcome the management problem of multi-source and massive science and education resources from a semantic level. However, the software-defined science and education resource model still faces a serious "knowledge navigation" problem when processing massive science and education resources, that is, users have difficulty in obtaining the resources they need from massive and multi-source distributed science and education resources.

[0003] In view of the knowledge navigation problem, a common solution is to construct a recommendation system based on a collaborative filtering method. However, in a multi-source distributed scenario, different science and education resources and the interaction information of users and resources are stored in different institutions, and due to privacy protection and knowledge security considerations, data sharing between different institutions is difficult, which brings difficulties to the science and education resource recommendation based on collaborative filtering.

[0004] In view of this, a recommendation scheme based on federated learning is proposed. Most of such schemes aggregate from the model level, which is difficult to effectively integrate the user preference features existing in different institutions and improve the recommendation accuracy. SUMMARY

[0005] To achieve the above object, the application adopts the following technical scheme:

[0006] In a first aspect, the application provides a software-defined science and education resource dynamic collaborative recommendation method, comprising:

[0007] Based on artificial intelligence knowledge entities and the relationship between entities, an artificial intelligence knowledge graph is constructed, science and education resource data of a science and education institution is acquired, knowledge entities in the science and education resource data are extracted, and connected to the artificial intelligence knowledge graph to obtain a science and education resource knowledge graph;

[0008] A user preference network model containing graph updating and time sequence feature fusion is constructed, local resource features / user features are extracted, and local resource features and user features of each client are obtained;

[0009] Based on the science and education resource knowledge graph, a local knowledge graph is constructed, and the local resource features of each client are used as initial features for model training until the model converges;

[0010] The graph neural network is used to predict user and resource information and recommends highly rated resources to each user.

[0011] Furthermore, based on AI knowledge entities and the relationships between entities, an AI knowledge graph is constructed, including:

[0012] Construct a public knowledge corpus in the field of artificial intelligence, and use entity recognition algorithms and relation extraction algorithms to extract knowledge entities and relationships between entities in the field of artificial intelligence, forming a knowledge graph in the field of artificial intelligence; specifically:

[0013] Corpus construction: Collect and label texts related to artificial intelligence to build a corpus;

[0014] Named entity recognition: On a given corpus, text features are extracted using the BERT language processing model, then the model is trained and its performance is evaluated on a validation set. The trained model is then applied to the complete corpus for named entity recognition.

[0015] Relation extraction: Extract relevant features, train the model using a recurrent neural network, evaluate the model performance on a validation set, and construct a knowledge graph by combining the extracted relations with the identified entities.

[0016] Furthermore, acquire science and education resource data from science and education institutions, extract knowledge entities from the data, connect them to an artificial intelligence knowledge graph, and obtain a science and education resource knowledge graph, including:

[0017] For each science and education institution, the metadata, text, video, image, and file types of data contained in its science and education resources are processed. The knowledge entities and their relationships are extracted, and then aligned and connected with the constructed artificial intelligence knowledge graph to build a software-defined science and education resource knowledge graph that integrates artificial intelligence knowledge points and science and education resources. Specifically:

[0018] Multimodal science and education resource feature extraction: Science and education resources of different modalities, including text, video, image and audio, are extracted separately. BERT is used to process text data, convolutional networks are used to process image and video data, etc., and Transformer is used to perform weighted fusion of the processed results to generate a comprehensive semantic feature representation of science and education resources. The category of each entity is output using this feature representation.

[0019] Entity matching from science and education resources to knowledge graphs: The learned features are measured by cosine similarity with the features of knowledge points in the knowledge graph, and the closest result is selected for linking, thus completing the knowledge link from science and education resources to entities in the knowledge graph.

[0020] Furthermore, a user preference network model incorporating graph updates and temporal feature fusion is constructed to extract local resource features / user features, obtaining local resource features and user features for each client, including:

[0021] For each client, a user preference network is constructed by graph modeling the "user-resource" preference relationship; the cross-entropy loss function is constructed using the preference relationship to train the model; and when the model converges, the local resource features and user features of each client are obtained.

[0022] Furthermore, a local knowledge graph is constructed based on the science and education resource knowledge graph. The local resource features of each client are used as initial features for model training until the model converges, including:

[0023] Model Construction: For science and education resource recommendation scenarios, a recommendation network is constructed based on graph convolutional neural networks;

[0024] Model training: Based on the constructed model, a loss function is built, and data is input into the model to calculate the loss and update the model parameters through backpropagation until the model converges.

[0025] Further, model building includes:

[0026] Graph update result extraction: extract the current time... The picture and The picture Compare and delete. The graph content at a given moment is used to obtain the updated graph. Subsequently Process it on;

[0027] Feature embedding module: At each layer, for each edge type Assign one Dimensional embedding And simultaneously use edge-type embeddings and node embeddings to compute attention scores:

[0028]

[0029] in Represents a node and nodes The type of the edges between them It is a learnable matrix used for transforming type embeddings;

[0030] Residual Link Module: Added pre-activated residual links for cross-layer node representation, assuming... It is the edge Attention weights It is an activation function, when the... When the feature dimensions of a layer change, a learnable linear transformation is used. The features are transformed, and the transformed features are then aggregated using residuals:

[0031]

[0032] Similarly, a residual module is also introduced for learning edge features, as detailed below:

[0033]

[0034] Multi-head attention module: Introduces a multi-head attention mechanism, leveraging... Each independent attention module learns the features, and the learned features are concatenated to obtain the final feature representation. The corresponding update rules are as follows:

[0035]

[0036]

[0037]

[0038] Final output layer The output of the layer is:

[0039]

[0040] Temporal feature fusion: Before the moment The features learned at each time point are respectively Then, a recurrent neural network is used to process the features, and the resulting features containing historical information are as follows:

[0041]

[0042] Preference information prediction: The output features are then input into a prediction module built on a multilayer perceptron, for each time step, the learned user features are... and resource characteristics The features are concatenated and then input into a multilayer perceptron to obtain a prediction score. as follows

[0043] .

[0044] Further, model training includes:

[0045] Constructing cross-entropy: During training, the dynamic changes of the graph are simulated to enhance the fit between the learned model and the real evolution. Specifically, for each time step involved in the training process... Construct corresponding training sets respectively ,for The included real labels and corresponding predictive label machines and The cross-entropy loss is constructed as follows:

[0046]

[0047] Model update: Next, the gradient is calculated using backpropagation of the neural network and the parameters are optimized. The stopping condition is evaluated. If the model has not converged, training continues until the model converges.

[0048] Secondly, this invention provides a dynamic collaborative recommendation system for science and education resources, including:

[0049] The science and education resource knowledge graph acquisition module is used to construct an artificial intelligence knowledge graph based on artificial intelligence knowledge entities and the relationships between entities; it acquires science and education resource data from science and education institutions, extracts knowledge entities from the science and education resource data, connects them to the artificial intelligence knowledge graph, and obtains the science and education resource knowledge graph.

[0050] The feature extraction module is used to construct a user preference network model that includes graph updates and temporal feature fusion, and to extract local resource features / user features to obtain the local resource features and user features of each client.

[0051] The model training module is used to build a local knowledge graph based on the science and education resource knowledge graph. The local resource features of each client are used as initial features for model training until the model converges.

[0052] The prediction output module is used to input users and resources into a graph neural network to make predictions and recommend high-rated resources to each user.

[0053] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs.

[0054] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs.

[0055] Compared with the prior art, the present invention has the following technical effects:

[0056] This invention utilizes open knowledge graphs as its core, cleverly breaking down the isolation of science and education resources and user preference information among different educational institutions, laying a solid foundation for building a unified and efficient distributed recommendation model. As a structured knowledge representation, knowledge graphs not only contain rich domain knowledge but also provide relationships between entities, making them a link connecting the resources of various educational institutions. Through this technology, institutions can share and utilize each other's resource information while protecting user privacy, thereby greatly expanding the data sources and coverage of the recommendation system and achieving true multi-source science and education resource recommendation.

[0057] Furthermore, this scheme uses knowledge graphs as a medium for feature propagation, combining them with graph convolutional neural networks to construct a powerful feature learning and propagation mechanism. Graph convolutional neural networks can deeply mine information within the graph structure, passing feature information from nodes to their neighboring nodes through convolution operations, thereby achieving feature aggregation and propagation. In this process, the knowledge graph, as the underlying graph structure, provides rich semantic information for resource features and user preference features, making feature learning and propagation more accurate and efficient. This mechanism not only effectively addresses the challenges posed by heterogeneous user preferences and resources but also improves the accuracy and robustness of recommendation models through semantic-level feature aggregation.

[0058] Furthermore, this technical solution fully considers the dynamic nature of user preferences and the efficiency requirements in scenarios involving the recommendation of massive amounts of science and education resources. By introducing a dynamic graph neural network, the solution achieves dynamic adjustment and optimization of the model structure. The dynamic graph neural network can adjust the feature representations of nodes and edges in real time according to changes in user behavior and resource updates, thereby keeping the model synchronized with user preferences and resource states. This dynamic adjustment mechanism avoids the model failure problem caused by changes in user preferences or resource updates in traditional recommendation systems, while also greatly improving the response speed and efficiency of the recommendation system. Since there is no need to retrain the model, this solution reduces system operation and maintenance costs and time costs while maintaining high efficiency.

[0059] In summary, this technical solution effectively addresses key issues in recommending multi-source science and education resources by utilizing open knowledge graphs, combining graph convolutional neural networks for feature propagation and aggregation, and introducing dynamic graph neural networks to achieve dynamic model adjustment. It provides strong technical support for building an efficient, accurate, and dynamic recommendation system. Attached Figure Description

[0060] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0061] The present invention will be further described below with reference to the accompanying drawings:

[0062] Example 1, please refer to Figure 1 This invention provides a software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs, including:

[0063] Based on artificial intelligence knowledge entities and the relationships between entities, an artificial intelligence knowledge graph is constructed; scientific and educational resource data from scientific and educational institutions is acquired, knowledge entities are extracted from the scientific and educational resource data, and connected to the artificial intelligence knowledge graph to obtain the scientific and educational resource knowledge graph.

[0064] A user preference network model incorporating graph updates and temporal feature fusion is constructed to extract local resource features / user features, thereby obtaining the local resource features and user features of each client.

[0065] A local knowledge graph is constructed based on the knowledge graph of science and education resources. The local resource features of each client are used as initial features for model training until the model converges.

[0066] The graph neural network is used to predict user and resource information and recommends highly rated resources to each user.

[0067] By constructing an artificial intelligence knowledge graph and integrating knowledge entities from science and education resource data of science and education institutions, a science and education resource knowledge graph containing rich domain knowledge and resource information has been formed. This graph not only provides a comprehensive knowledge background for the recommendation system, but also reveals the potential connections between resources through the relationships between entities, thereby improving the accuracy and relevance of recommendations.

[0068] The dynamism and temporal nature of user preference network models enhance the personalization of recommendations:

[0069] The constructed user preference network model not only considers users' current preferences but also captures dynamic changes in user preferences and historical behavioral patterns through graph updates and temporal feature fusion mechanisms. This enables the recommender system to more accurately understand users' personalized needs and provide resource recommendations that match users' current interests and historical preferences, enhancing the personalization and timeliness of recommendations.

[0070] This method allows each client to train the model based on its local knowledge graph and resource features until the model converges. This distributed training approach not only reduces dependence on a central server but also improves the efficiency and scalability of model training. Simultaneously, by inputting user and resource information into a graph neural network for prediction and quickly recommending highly-rated resources to each user, the real-time performance and responsiveness of the recommendation system are ensured.

[0071] As a core component of recommender systems, knowledge graphs possess powerful semantic understanding capabilities, enabling recommender systems to gain a deeper understanding of resource content and user intent. This not only improves the accuracy of recommendations but also enhances the interpretability and credibility of the recommendations, making it easier for users to accept and trust the system's suggestions.

[0072] Example 2: This invention provides a software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs, specifically including:

[0073] The software-defined collaborative recommendation algorithm for science and education resources based on knowledge graphs is applicable to systems containing... Different science and education institutions For each institution ,Include Individual users Individual science and education resources and the relationship between users and science and education resources. In addition, the system contains a knowledge graph in the field of artificial intelligence, which is composed of knowledge points related to artificial intelligence. The main purpose of the software-defined multi-source collaborative recommendation algorithm for science and education resources based on knowledge graphs is to learn the characteristics of users and resources from the user preference relationship network of science and education resources maintained by different science and education institutions, and to fuse them with artificial intelligence knowledge graphs, thereby extracting more comprehensive user and resource characteristics to achieve more accurate resource recommendations.

[0074]

[0075] 1. Public Knowledge Graph Construction: Construct a public knowledge corpus in the field of artificial intelligence. Utilize entity recognition algorithms and relation extraction algorithms to extract knowledge entities and relationships between entities in the field of artificial intelligence, forming a knowledge graph for the artificial intelligence domain. ;

[0076] Local science and education resource links: for each science and education institution The system processes various types of data, including metadata, text, videos, images, and files, contained in its science and education resources. It extracts the knowledge entities and their relationships, aligns and connects these with a constructed artificial intelligence knowledge graph, and builds a software-defined science and education resource knowledge graph that integrates artificial intelligence knowledge points with science and education resources. ;

[0077] Local resource / user feature extraction based on graph neural networks: In each client, a user preference network is constructed by modeling the "user-resource" preference relationship as a graph. The cross-entropy loss function is constructed using preference relationships to train the model. Once the model converges, the local resource features and user features of each client are obtained.

[0078] Cross-entity feature propagation based on knowledge graphs: various entities Utilize local science and education resources and shared artificial intelligence knowledge graph Building a local knowledge graph Using the science and education resource features learned in step (3) as initial features, they are input into a convolutional neural network for feature learning. For each feature... Each entity in , Learn local features Subsequently, all institutions shared all of the data. Subsequently, each client uses an attention mechanism to fuse the knowledge point features, obtaining a comprehensive set of knowledge point features. ;

[0079] Local feature update: given Update local resource characteristics and combine the updated resource characteristics with... Input a graph neural network to update local user and resource features;

[0080] Model training: Repeat the process until the model converges;

[0081] Recommendation of science and education resources based on federated collaborative filtering: Input users and resources into a graph neural network to make predictions, and recommend resources with higher ratings to each user.

[0082] The knowledge graph construction described in section 1 mainly includes the following processes:

[0083] Corpus construction: Collect and label texts related to artificial intelligence to build a corpus.

[0084] Named entity recognition: On a given corpus, text features are extracted using the BERT language processing model. The model is then trained and its performance is evaluated on a validation set. Parameters are adjusted to improve the results. The trained model is then applied to the complete corpus for named entity recognition.

[0085] Relation extraction: Extract relevant features, train the model using a recurrent neural network, evaluate the model performance on a validation set, and construct a knowledge graph by combining the extracted relations with the identified entities.

[0086] The science and education resource links shown in Figure 1 mainly include the following steps:

[0087] Multimodal science and education resource feature extraction: Science and education resources of different modalities, including text, video, image, and audio, are extracted separately. For example, BERT is used to process text data, convolutional networks are used to process image and video data, and Transformer is used to weight and fuse the processed results to generate a comprehensive semantic feature representation of science and education resources. This feature representation is then used to output the category of each entity.

[0088] Entity matching from science and education resources to knowledge graphs: The learned features are compared with the features of knowledge points in the knowledge graph using cosine similarity measurement, and the closest result is selected for linking, thus completing the process of matching science and education resources to knowledge graphs. Knowledge links for entities in the middle.

[0089] The resource / user feature extraction based on graph neural networks, as shown in Figure 1, includes the following steps:

[0090] Model Construction: For science and education resource recommendation scenarios, a recommendation network is constructed based on graph convolutional neural networks;

[0091] Model training: Based on the constructed model, a loss function is built, and data is input into the model to calculate the loss and update the model parameters through backpropagation until the model converges;

[0092] Model building mainly includes the following modules:

[0093] Graph update result extraction: extract the current time... The picture and The picture Compare and delete. The graph content at a given moment is used to obtain the updated graph. Subsequently Process it on;

[0094] Feature embedding module: At each layer, for each edge type Assign one Dimensional embedding And simultaneously use edge-type embeddings and node embeddings to compute attention scores:

[0095]

[0096] in Represents a node and nodes The type of the edges between them It is a learnable matrix used for transforming type embeddings;

[0097] Residual Link Module: Added pre-activated residual links for cross-layer node representation, assuming... It is the edge Attention weights It is an activation function, when the... When the feature dimensions of a layer change, the scheme uses a learnable linear transformation. The features are transformed, and the transformed features are then aggregated using residuals:

[0098]

[0099] Similarly, a residual module is also introduced for learning edge features, as detailed below:

[0100]

[0101] Multi-head attention module: To further enhance the model's learning ability, a multi-head attention mechanism is introduced, leveraging... Each independent attention module learns the features, and the learned features are concatenated to obtain the final feature representation. The corresponding update rules are as follows:

[0102]

[0103]

[0104]

[0105] Final output layer The output of the layer is:

[0106]

[0107] Temporal feature fusion: Before the moment The features learned at each time point are respectively Then, a recurrent neural network is used to process the features, and the resulting features containing historical information are as follows:

[0108]

[0109] Preference information prediction: The output features are then input into a prediction module built on a multilayer perceptron, for each time step, the learned user features are... and resource characteristics The features are concatenated and then input into a multilayer perceptron to obtain a prediction score. as follows

[0110]

[0111] Model training mainly includes the following steps:

[0112] Constructing cross-entropy: During training, the dynamic changes of the graph are simulated to enhance the fit between the learned model and the real evolution. Specifically, for each time step involved in the training process... Construct corresponding training sets respectively ,for The included real labels and corresponding predictive label machines and The cross-entropy loss is constructed as follows:

[0113]

[0114] Model update: Next, the gradient is calculated using backpropagation of the neural network and the parameters are optimized. The stopping condition is evaluated. If the model does not converge, the process is repeated until the model converges.

[0115] This invention utilizes open knowledge graphs as a barrier to connect science and education resources and user preference information stored by different educational institutions, which can effectively enable the learning of distributed recommendation models and achieve privacy-preserving multi-source science and education resource recommendations.

[0116] This invention uses knowledge graphs as feature propagation media and combines them with graph convolutional neural networks to achieve effective propagation of resource features and user preference features between different institutions. It also achieves feature aggregation at the semantic level, which can better resist the performance degradation problem of recommendation models caused by heterogeneous user preferences and resources, and thus effectively improve model accuracy.

[0117] This invention comprehensively considers the dynamic nature of user preferences and the efficiency requirements in scenarios involving the recommendation of massive amounts of science and education resources. It utilizes dynamic graph neural networks to achieve dynamic adjustment of the model, enabling it to fit dynamically changing user preference information without requiring retraining.

[0118] In another embodiment of the present invention, a software-defined dynamic collaborative recommendation system for science and education resources based on knowledge graphs is provided, which can be used to implement the above-mentioned software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs. Specifically, the system includes:

[0119] The science and education resource knowledge graph acquisition module is used to construct an artificial intelligence knowledge graph based on artificial intelligence knowledge entities and the relationships between entities; it acquires science and education resource data from science and education institutions, extracts knowledge entities from the science and education resource data, connects them to the artificial intelligence knowledge graph, and obtains the science and education resource knowledge graph.

[0120] The feature extraction module is used to construct a user preference network model that includes graph updates and temporal feature fusion, and to extract local resource features / user features to obtain the local resource features and user features of each client.

[0121] The model training module is used to build a local knowledge graph based on the science and education resource knowledge graph. The local resource features of each client are used as initial features for model training until the model converges.

[0122] The prediction output module is used to input users and resources into a graph neural network to make predictions and recommend high-rated resources to each user.

[0123] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0124] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs.

[0125] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the software-defined dynamic collaborative recommendation method for knowledge graph-based science and education resources in the above embodiments.

[0126] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0128] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs, characterized in that: include: Construct an AI knowledge graph based on AI knowledge entities and the relationships between entities; Acquire science and education resource data from science and education institutions, extract knowledge entities from the science and education resource data, connect them to an artificial intelligence knowledge graph, and obtain a science and education resource knowledge graph; A user preference network model incorporating graph updates and temporal feature fusion is constructed to extract local resource features / user features, thereby obtaining the local resource features and user features of each client. A local knowledge graph is constructed based on the knowledge graph of science and education resources. The local resource features of each client are used as initial features for model training until the model converges. The graph neural network is used to predict user and resource information and recommends high-rated resources to each user. A user preference network model incorporating graph updates and temporal feature fusion is constructed to extract local resource features / user features, obtaining local resource features and user features for each client, including: In each client, a user preference network is constructed by graph modeling the "user-resource" preference relationship; the cross-entropy loss function is constructed using the preference relationship to train the model; when the model converges, the local resource features and user features of each client are obtained. A local knowledge graph is constructed based on a science and education resource knowledge graph. The local resource features of each client are used as initial features for model training until the model converges, including: Model Construction: For science and education resource recommendation scenarios, a recommendation network is constructed based on graph convolutional neural networks; Model training: Based on the constructed model, a loss function is built, and data is input into the model to calculate the loss and update the model parameters through backpropagation until the model converges; Model building includes: Graph update result extraction: extract the current time... The picture and The picture Compare and delete. The graph content at a given moment is used to obtain the updated graph. Subsequently Process it on; Feature embedding module: At each layer, for each edge type Assign one Dimensional embedding And simultaneously use edge-type embeddings and node embeddings to compute attention scores: in Represents a node and nodes The type of the edges between them It is a learnable matrix used for transforming type embeddings; Residual Link Module: Added pre-activated residual links for cross-layer node representation, assuming... It is the edge Attention weights It is an activation function, when the... When the feature dimensions of a layer change, a learnable linear transformation is used. The features are transformed, and the transformed features are then aggregated using residuals. Similarly, a residual module is also introduced for learning edge features, as detailed below: Multi-head attention module: Introduces a multi-head attention mechanism, leveraging... Each independent attention module learns the features, and the learned features are concatenated to obtain the final feature representation. The corresponding update rules are as follows: Final output layer The output of the layer is: Temporal feature fusion: Before the moment The features learned at each time point are respectively Then, a recurrent neural network is used to process the features, and the resulting features containing historical information are as follows: Preference information prediction: The output features are then input into a prediction module built on a multilayer perceptron, for each time step, the learned user features are used. and resource characteristics The features are concatenated and then input into a multilayer perceptron to obtain a prediction score. as follows ； Model training includes: Constructing cross-entropy: During training, the dynamic changes of the graph are simulated to enhance the fit between the learned model and the real evolution. Specifically, for each time step involved in the training process... Construct corresponding training sets respectively ,for The included real labels and corresponding predictive label machines and The cross-entropy loss is constructed as follows: Model update: Next, the gradient is calculated using backpropagation of the neural network and the parameters are optimized. The stopping condition is evaluated. If the model has not converged, training continues until the model converges.

2. The software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs according to claim 1, characterized in that, Based on AI knowledge entities and the relationships between entities, an AI knowledge graph is constructed, including: Construct a public knowledge corpus in the field of artificial intelligence, and use entity recognition algorithms and relation extraction algorithms to extract knowledge entities and relationships between entities in the field of artificial intelligence, forming a knowledge graph in the field of artificial intelligence; specifically: Corpus construction: Collect and label texts related to artificial intelligence to build a corpus; Named entity recognition: On a given corpus, text features are extracted using the BERT language processing model, then the model is trained and its performance is evaluated on a validation set. The trained model is then applied to the complete corpus for named entity recognition. Relation extraction: Extract relevant features, train the model using a recurrent neural network, evaluate the model performance on a validation set, and construct a knowledge graph by combining the extracted relations with the identified entities.

3. The software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs according to claim 1, characterized in that, Acquire science and education resource data from science and education institutions, extract knowledge entities from the data, connect them to an artificial intelligence knowledge graph, and obtain a science and education resource knowledge graph, including: For each science and education institution, the metadata, text, video, image, and file types of data contained in its science and education resources are processed. The knowledge entities and their relationships are extracted, and then aligned and connected with the constructed artificial intelligence knowledge graph to build a software-defined science and education resource knowledge graph that integrates artificial intelligence knowledge points and science and education resources. Specifically: Multimodal science and education resource feature extraction: Science and education resources of different modalities, including text, video, image and audio, are extracted separately. BERT is used to process text data, convolutional networks are used to process image and video data, etc., and Transformer is used to perform weighted fusion of the processed results to generate a comprehensive semantic feature representation of science and education resources. The category of each entity is output using this feature representation. Entity matching from science and education resources to knowledge graphs: The learned features are measured by cosine similarity with the features of knowledge points in the knowledge graph, and the closest result is selected for linking, thus completing the knowledge link from science and education resources to entities in the knowledge graph.

4. A software-defined dynamic collaborative recommendation system for science and education resources based on knowledge graphs, characterized in that: include: The science and education resource knowledge graph acquisition module is used to construct an artificial intelligence knowledge graph based on artificial intelligence knowledge entities and the relationships between entities; Acquire science and education resource data from science and education institutions, extract knowledge entities from the science and education resource data, connect them to an artificial intelligence knowledge graph, and obtain a science and education resource knowledge graph; The feature extraction module is used to construct a user preference network model that includes graph updates and temporal feature fusion, and to extract local resource features / user features to obtain the local resource features and user features of each client. The model training module is used to build a local knowledge graph based on the science and education resource knowledge graph. The local resource features of each client are used as initial features for model training until the model converges. The prediction output module is used to input users and resources into a graph neural network to make predictions and recommend high-rated resources to each user. A user preference network model incorporating graph updates and temporal feature fusion is constructed to extract local resource features / user features, obtaining local resource features and user features for each client, including: In each client, a user preference network is constructed by graph modeling the "user-resource" preference relationship; the cross-entropy loss function is constructed using the preference relationship to train the model; when the model converges, the local resource features and user features of each client are obtained. A local knowledge graph is constructed based on a science and education resource knowledge graph. The local resource features of each client are used as initial features for model training until the model converges, including: Model Construction: For science and education resource recommendation scenarios, a recommendation network is constructed based on graph convolutional neural networks; Model training: Based on the constructed model, a loss function is built, and data is input into the model to calculate the loss and update the model parameters through backpropagation until the model converges; Model building includes: Graph update result extraction: extract the current time... The picture and The picture Compare and delete. The graph content at a given moment is used to obtain the updated graph. Subsequently Process it on; Feature embedding module: At each layer, for each edge type Assign one Dimensional embedding And simultaneously use edge-type embeddings and node embeddings to compute attention scores: in Represents a node and nodes The type of the edges between them It is a learnable matrix used for transforming type embeddings; Residual Link Module: Added pre-activated residual links for cross-layer node representation, assuming... It is the edge Attention weights It is an activation function, when the... When the feature dimensions of a layer change, a learnable linear transformation is used. The features are transformed, and the transformed features are then aggregated using residuals. Similarly, a residual module is also introduced for learning edge features, as detailed below: Multi-head attention module: Introduces a multi-head attention mechanism, leveraging... Each independent attention module learns the features, and the learned features are concatenated to obtain the final feature representation. The corresponding update rules are as follows: Final output layer The output of the layer is: Temporal feature fusion: Before the moment The features learned at each time point are respectively Then, a recurrent neural network is used to process the features, and the resulting features containing historical information are as follows: Preference information prediction: The output features are then input into a prediction module built on a multilayer perceptron, for each time step, the learned user features are used. and resource characteristics The features are concatenated and then input into a multilayer perceptron to obtain a prediction score. as follows ； Model training includes: Constructing cross-entropy: During training, the dynamic changes of the graph are simulated to enhance the fit between the learned model and the real evolution. Specifically, for each time step involved in the training process... Construct corresponding training sets respectively ,for The included real labels and corresponding predictive label machines and The cross-entropy loss is constructed as follows: Model update: Next, the gradient is calculated using backpropagation of the neural network and the parameters are optimized. The stopping condition is evaluated. If the model has not converged, training continues until the model converges.

5. A computer 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 computer program, it implements the steps of the software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs as described in any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the software-defined dynamic collaborative recommendation method for science and education resources based on knowledge graphs as described in any one of claims 1 to 3.

Images