Heterogeneous graph neural network-based search model establishment method and system and application

Abstract

The invention discloses a search model establishment method and system based on a heterogeneous graph neural network and application, and belongs to the technical field of artificial intelligence and graph neural networks. The search model establishment method comprises the following steps: collecting meta-paths related to a target node to form an initial meta-path set; generating a meta-path redundancy graph by utilizing the meta-paths of which the dependency degree between every two in the initial meta-path set is higher than a preset threshold value; searching an optimal non-adjacent meta-path combination from the meta-path redundancy graph as an initial non-redundant meta-path combination; according to meta-paths and meta-path strength in the initial non-redundant meta-path combination, optimizing architecture parameters and network weights of the initial non-redundant meta-path combination to obtain an optimal non-redundant meta-path combination; establishing a meta-path aggregation search space as a search model for the optimal non-redundant meta-path combination; and optimizing and training a search model by using a micromodular architecture search strategy. According to the method, the performance and efficiency of representing the learning task by the heterogeneous graph are improved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence and graph neural network technology, and more specifically, relates to a method, system and application for establishing a search model based on heterogeneous graph neural network. Background Art

[0002] Heterogeneous graph neural networks, as an important tool for processing complex heterogeneous information networks, have demonstrated significant value in fields such as social network analysis and knowledge graph reasoning by mining high-order semantic connections between different types of nodes. The core mechanism of heterogeneous graph neural networks relies on the structured semantic modeling of meta-paths, which can transform multi-hop interactions between heterogeneous nodes into computable graph topological features. However, existing technology systems face multi-dimensional challenges in the design and optimization of key links.

[0003] In the context of recommender systems, node types such as users and products form a complex e-commerce network, which can be abstracted as a heterogeneous graph. In this graph, each node has a corresponding type, and each edge has a clear relationship type, fully representing the complex relationships between users, orders, categories, and products in e-commerce scenarios. Heterogeneous graphs contain a variety of complex meta-paths, such as user-order-category-product-user, user-product-user, and user-product-type. This complexity requires recommender systems to process massive amounts of product information and diverse user behaviors. However, existing technologies are inefficient in handling this complexity and consume significant computational resources. Traditional meta-path selection methods, such as random sampling or exhaustive screening, face exponentially growing meta-path combination space, resulting in rapidly increasing computational resource consumption and difficulty in efficiently finding the optimal meta-path combination. Furthermore, a large number of invalid or low-value meta-paths introduce semantic noise, interfering with the accurate extraction of user preferences and product association features. This not only hinders further performance improvements in recommender systems but also directly impacts the accuracy and recall of recommendations, thereby reducing user experience and commercial value.

[0004] Similarly, similar challenges exist in other fields. For example, in a heterogeneous graph consisting of four node types: author-paper-term-author, researchers aim to mine relationships such as academic collaboration and knowledge inheritance. However, current feature aggregation frameworks typically use static weight assignment to integrate multi-path features. This fixed aggregation strategy ignores the differentiated semantic information needs of different academic mining tasks (such as academic influence assessment and research direction prediction). This rigid architecture is particularly inadequate in the context of interdisciplinary research integration or the dynamic evolution of academic fields. It can easily lead to the loss of key academic features or the over-amplification of noise features, making it difficult for mining results to accurately reflect the true state of academic relationships. This not only hinders the improvement of academic knowledge graphs, but also limits the in-depth development of academic research, making it difficult to meet the increasingly complex needs of modern academic research.

[0005] In summary, existing search methods based on heterogeneous graph neural networks have problems of insufficient accuracy and low efficiency. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a method, system and application for establishing a search model based on heterogeneous graph neural networks, so as to solve the problems of insufficient accuracy and low efficiency of the existing search methods based on heterogeneous graph neural networks.

[0007] Based on the above-mentioned purpose, the first aspect of the present invention provides a method for establishing a search model based on a heterogeneous graph neural network, including: collecting meta-paths related to the target node to form an initial meta-path set; using the meta-paths in the initial meta-path set whose pairwise dependencies are higher than a preset threshold to generate a meta-path redundancy graph; searching for the optimal non-adjacent meta-path combination from the meta-path redundancy graph as the initial non-redundant meta-path combination; optimizing the architecture parameters and network weights of the initial non-redundant meta-path combination according to the meta-paths and meta-path strengths in the initial non-redundant meta-path combination to obtain the optimal non-redundant meta-path combination; establishing a meta-path aggregation search space for the optimal non-redundant meta-path combination as a search model; and optimizing and training the search model using a differentiable modular architecture search strategy.

[0008] Optionally, the dependency is: ; in, Metapath and metapath The dependence between 、 Metapath , meta path The set of nodes covered, 、 Metapath , meta path The set of covered edges.

[0009] Optionally, the meta-path strength is: ; in, Metapath The meta-path strength of 、 and meta-path respectively , meta path The associated schema parameters, is the total number of meta-paths in the initial non-redundant meta-path combination, 、 Metapath , meta path The size of the node set covered, 、 Metapath , meta path The size of the edge set covered, is the temperature parameter.

[0010] Optionally, optimizing the architecture parameters and network weights of the initial non-redundant meta-path combination specifically includes: alternately updating the architecture parameters and the network weights until optimal architecture parameters and network weights are obtained.

[0011] Optionally, the meta-path aggregation search space can realize one or more of message passing, feature aggregation and meta-path fusion; the message passing uses direct aggregation method, average aggregation method and / or multi-layer perceptron transformation method to transmit node information in the meta-path; the feature aggregation uses average pooling method, maximum pooling method and / or attention mechanism to aggregate the features after message passing; the meta-path fusion uses average pooling method, single-head attention mechanism and / or multi-head attention mechanism to fuse the features after feature aggregation, and output the meta-path fused features.

[0012] Optionally, the search model is optimized and trained using a differentiable modular architecture search strategy, specifically including: dividing the meta-path aggregation process in the search model into multiple modules, each module containing a set of candidate operations; using a Softmax weighted hypernetwork to convert discrete architecture choices into continuous probabilities, and jointly optimizing the weight parameters and model weights in the search model through gradient descent.

[0013] The second aspect of the present invention provides a search method based on a heterogeneous graph neural network, comprising: using the optimized search model in the above-mentioned method to search for comprehensive feature information related to the target node.

[0014] The third aspect of the present invention provides a search model establishment system based on a heterogeneous graph neural network, including: a collection module for collecting meta-paths related to a target node to form an initial meta-path set; a generation and search module for generating a meta-path redundancy graph using meta-paths in the initial meta-path set whose dependencies are higher than a preset threshold; finding the optimal non-adjacent meta-path combination from the meta-path redundancy graph as the initial non-redundant meta-path combination; a first optimization module for optimizing the architecture parameters and network weights of the initial non-redundant meta-path combination based on the meta-paths and meta-path strengths in the initial non-redundant meta-path combination to obtain the optimal non-redundant meta-path combination; an establishment module for establishing a meta-path aggregation search space for the optimal non-redundant meta-path combination as a search model; and a second optimization module for optimizing and training the search model using a differentiable modular architecture search strategy.

[0015] A fourth aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described above when executing the program.

[0016] A fifth aspect of the present invention provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method described above.

[0017] Compared with the existing technology, the advantages of the present invention include: providing a search model establishment method based on heterogeneous graph neural network, using the meta-paths in the initial meta-path set whose pairwise dependencies are higher than a preset threshold to generate a meta-path redundancy graph, effectively reducing information redundancy and accurately selecting the most appropriate meta-path combination; with the help of a differentiable modular architecture search strategy, comprehensively mining the structural and semantic information of the heterogeneous graph, significantly improving the performance and applicability of the search model, thereby improving the accuracy and efficiency of the search results; experimental results show that this method is superior to existing advanced methods and can dynamically adjust the meta-path aggregation strategy to adapt to different task requirements, thereby achieving higher accuracy and efficiency in heterogeneous graph representation learning tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a method for establishing a search model based on a heterogeneous graph neural network provided in an embodiment of the present invention.

[0019] Figure 2 Implementation provided by the embodiment of the present invention Figure 1 System framework diagram of the method shown.

[0020] Figure 3 A block diagram of a search model establishment system based on a heterogeneous graph neural network provided in an embodiment of the present invention.

[0021] Figure 4 A schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In view of the shortcomings of the prior art, the inventors of this case, after long-term research and extensive practice, have proposed the technical solution of the present invention. The following will further explain this technical solution, its implementation process and principles.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0024] In addition, in the description of the present invention, it should be understood that the terms "upper", "lower", "inside", "outside", "horizontal", "vertical", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0025] Throughout this specification, references to "one embodiment," "an embodiment," "the embodiment," and the like indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Throughout this specification, the schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0026] The embodiment of the present invention provides a method for establishing a search model based on a heterogeneous graph neural network. Figure 1 The method for establishing a search model based on a heterogeneous graph neural network includes steps S100 to S500.

[0027] Step S100 : collecting meta-paths related to the target node to form an initial meta-path set.

[0028] Collect all possible meta-paths related to the target node to form an initial meta-path set. The target node is the node type to be classified or predicted. Each meta-path is composed of different types of nodes and relationships in a certain order, such as a path like author-paper-term-author.

[0029] Step S200 , generating a meta-path redundancy graph using meta-paths in the initial meta-path set whose dependencies between each other are higher than a preset threshold; and finding the optimal non-adjacent meta-path combination in the meta-path redundancy graph as the initial non-redundant meta-path combination.

[0030] For each meta-path in the initial meta-path set, the set of nodes and edges it covers is calculated. For any two meta-paths, the dependency between them is calculated by comparing the set of nodes and edges they cover. Specifically, the sum of the ratios of the intersection and union of the two meta-paths is calculated. This quantifies the dependencies between different meta-paths and identifies semantically similar or redundant meta-paths.

[0031] The initial meta-path set can be expressed as . Preferably, for any two meta-paths and , the dependency between them is: ; in, Metapath and metapath The dependence between 、 Metapath , meta path The set of nodes covered, 、 Metapath , meta path The set of covered edges.

[0032] Preferably, the process of generating the meta-path redundancy graph is as follows: if the meta-path and Dependence between Above the preset threshold δ, in the meta-path redundancy graph Add an edge to connect them, marking the highly dependent meta-path combination; select the unselected nodes through the greedy algorithm, add them to the meta-path combination, and delete the node and its adjacent nodes until all nodes are processed. is a node set, representing the set of all meta-paths; is an edge set, representing the redundant relationship between meta-paths.

[0033] Further preferably, a greedy algorithm is used to find the optimal non-adjacent meta-path combination from the meta-path redundancy graph, that is, a group of meta-paths without direct dependencies are selected as the initial non-redundant meta-path combination to reduce resource consumption in subsequent calculations.

[0034] Step S300 , optimizing the architecture parameters and network weights of the initial non-redundant meta-path combination according to the meta-paths and meta-path strengths in the initial non-redundant meta-path combination to obtain an optimal non-redundant meta-path combination.

[0035] Preferably, the meta-path strength is: ; in, Metapath The meta-path strength of 、 and meta-path respectively , meta path The associated schema parameters, is the total number of meta-paths in the initial non-redundant meta-path combination, 、 Metapath , meta path The size of the node set covered, 、 Metapath , meta path The size of the edge set covered, is the temperature parameter. Architecture parameter .

[0036] Preferably, in one embodiment of the present invention, optimizing the architecture parameters and network weights of the initial non-redundant meta-path combination specifically includes: alternately updating the architecture parameters and network weights until the optimal architecture parameters and network weights are obtained.

[0037] The specific process of the optimization operation in step S300 is exemplarily described below.

[0038] Use multilayer perceptron to combine the initial non-redundant meta-paths The features of the meta-path are transformed and aggregated to obtain the updated node representation. The update formula is: ; in, Target node Update representation of ; is the projection function; Represents the projection function The weight parameter in is the network weight; Metapath The feature matrix of .

[0039] Adopting alternating update strategy to optimize architecture parameters and network weights That is, one parameter is fixed, and the other parameter is optimized, and the process is alternated until the most suitable network architecture is found, which can be represented as: ; ; wherein, and represent the training set loss and the validation set loss, respectively, and t e {1, 2, …, T} represents the iteration number. In each iteration, the architecture parameters are first fixed to optimize the network weights, and then the updated network weights are fixed to optimize the architecture parameters.

[0040] Without loss of generality, when dealing with node classification tasks, the training set loss and the validation set loss can be represented as: ; wherein, is the number of target nodes, is the number of classes, is the true label, is the probability of the cth class. is the predicted probability vector of the target node , and the cth element of corresponds to the probability of the cth class . is obtained by converting into a probability distribution of different classes through the softmax function of the output layer: ; wherein, represents a vector with all elements being 1, and T represents transposition, is a component of , representing the feature intensity or predicted score of the target node on different classes.

[0041] Step S400, establishing an elementary path aggregation search space as a search model for the optimal non-redundant elementary path combination.

[0042] The elementary path aggregation search space is designed to contain multiple functions, and preferably, the elementary path aggregation search space can realize one or more of message passing, feature aggregation, and elementary path fusion. The message passing function is used to pass information between the nodes of the elementary path; the feature aggregation function is used to integrate the features of different instances under the same elementary path; and the elementary path fusion function is used to fuse the features on different elementary paths.

[0043] Preferably, the message transmission uses direct aggregation method, average aggregation method and / or multi-layer perceptron transformation method to transmit the node information in the meta-path to obtain the feature after the message transmission. .

[0044] The direct aggregation method only considers the starting point and end point of the path and ignores the information of the intermediate nodes in the path. , the features after direct aggregation for: ; in, Express about and The corresponding adjacency matrix is, Represents a meta path Previous i Node types, Represents a meta path Nodes on represents the optimal non-redundant meta-path combination, Indicates the length of the corresponding meta-path, Represents a direct aggregation operation.

[0045] The average aggregation method generates new aggregate features by averaging the features of all nodes on the path element by element. The aggregated features for: ; The multi-layer perceptron transformation method performs multi-layer nonlinear transformation on node features through multi-layer perceptron transformation, which improves the model's expressive power and ability to capture complex semantic relationships. The aggregated features for: ; Preferably, feature aggregation uses average pooling method, maximum pooling method and / or attention mechanism to aggregate the features after message transmission to obtain the aggregated features .

[0046] The average pooling method averages all instance features under the same meta-path element by element, which can effectively integrate the information of all instances and obtain a comprehensive feature representation. for: ; in, is the number of meta-path instances, is the feature of the i-th meta-path instance.

[0047] The maximum pooling method extracts the maximum value from all instance features under the same meta-path to highlight the most important features. for: ; The attention mechanism learns the normalized importance weight of each meta-path instance and performs weighted summation to model and flexibly capture the different contributions of different instances to the target node. The aggregated features for: ; in, represents the neighbors based on meta-path, Indicates that node u is in the meta-path The attention weight of node v is given below.

[0048] Preferably, the meta-path fusion uses the average pooling method, the single-head attention mechanism and / or the multi-head attention mechanism to fuse the features after feature aggregation, and outputs the meta-path fused features. .

[0049] The average pooling method takes into account the information of different meta-paths and obtains a comprehensive feature representation by averaging each element. The fused features for: ; Among them, K is the number of meta-paths, It is the feature representation of the target node v after meta-path fusion.

[0050] The single-head attention mechanism dynamically adjusts the feature aggregation method according to the importance of the meta-path. It takes into account the different contributions of different meta-paths to the target node, weighs and sums the relevant meta-paths, and the fused features for: ; in, is the normalized attention score.

[0051] The multi-head attention mechanism is extended to a multi-head version to solve the high variance problem caused by heterogeneity in large-scale heterogeneous graphs. The specific implementation is to perform R independent attention mechanisms and then connect their outputs to fused features. for: ; in, is the Softmax activation function, For the An attention head for The normalized importance weight of After feature aggregation Meta-path features, is the total number of meta-paths in the optimal non-redundant meta-path combination, is the number of independent attention mechanisms.

[0052] Step S500: Optimize and train the search model using a differentiable modular architecture search strategy.

[0053] Preferably, in one embodiment of the present invention, step S500 specifically includes: dividing the meta-path aggregation process in the search model into multiple modules, each module containing a set of candidate operations; using a Softmax weighted super network to convert discrete architecture choices into continuous probabilities, and jointly optimizing the weight parameters and model weights in the search model through gradient descent.

[0054] The differentiable modular architecture search strategy relaxes discrete choices through Softmax weighting, converting architecture selection into continuous probability. The weight parameters in the search model are jointly optimized by gradient descent. , γ and model weights to determine the weight of each operation. After training, the operation with the highest weight is selected to build a deterministic architecture. The formula can be expressed as: ; in, Represents the message passing module ( =1) or feature aggregation module ( =2), Indicates the corresponding operation, is the weight vector corresponding to the operation, Indicates different meta-path types, Represents the result of the weighted operation combination of the softmax function. In the meta-path fusion module, the corresponding operation weight calculation formula is as follows: ; in, is the weight vector corresponding to the operation at this module. Through modular differentiable search, the weight parameter { , } and model weights are jointly optimized to determine the weight of each operation.

[0055] Select the optimal operations for the three stages of message passing, feature aggregation, and meta-path fusion 、 、 , forming a deterministic architecture, expressed as: ; in, Contains the comprehensive feature information of the target node v. Taking the recommendation system scenario as an example, The information contained is obtained by modeling and aggregating nodes such as users, orders, categories, and products and the complex relationships between them. It can be used for various downstream tasks, such as node classification (predicting the category of nodes, such as the user's interest category in a certain product), link prediction (predicting whether a user is likely to purchase a certain product), etc.

[0056] The present invention provides a method for establishing a search model based on a heterogeneous graph neural network, which constructs a search strategy for identifying meta-path dependencies; generates an initial meta-path set, constructs a meta-path redundancy graph based on the dependencies between meta-paths, and redundantly marks highly dependent meta-paths; uses a greedy algorithm to find an optimal set of non-adjacent nodes, i.e., a non-redundant meta-path combination, in the meta-path redundancy graph; defines meta-path strength, and aggregates node information based on the selected meta-path and its strength to update the node representation; dynamically finds the most suitable network architecture by alternately optimizing architecture parameters and network weights; designs a meta-path aggregation search space that includes multiple functions, and adopts a micro-modular architecture search strategy to achieve personalized meta-path information aggregation. The present invention optimizes the meta-path sampling strategy by constructing a meta-path redundancy graph, adopts a micro-modular architecture search strategy to identify an adaptive meta-path aggregation strategy, comprehensively mines the structural and semantic information of the heterogeneous graph, improves information aggregation efficiency, and reduces computational redundancy, effectively improving the performance and efficiency of heterogeneous graph representation learning tasks, and is suitable for scenarios such as social network analysis, recommendation systems, and academic relationship mining.

[0057] For recommendation system application scenarios, the present invention solves the dual dilemma of existing automated meta-path selection mechanisms in terms of computational efficiency and semantic noise. By sampling non-redundant meta-path combinations and optimizing network architecture through a greedy algorithm, computing resource consumption is reduced, while semantic noise interference is reduced, significantly improving the model's ability to identify user preferences and product-related features, and improving the accuracy and recall rate of recommendations. In particular, when processing large-scale user and product data and multimodal user behavior data, the advantages are more obvious. In the field of academic relationship mining, the present invention overcomes the drawback of the lack of dynamic adaptability of existing feature aggregation frameworks. According to the requirements of different academic mining tasks, such as academic influence evaluation, research direction prediction, academic cooperation recommendation, etc., the dynamic architecture search strategy of the present invention can flexibly adjust the feature aggregation method to avoid the loss of key academic features or excessive amplification of noise features, so that the mining results can more accurately reflect the true situation of academic relationships.

[0058] The advantages of the search model establishment method based on heterogeneous graph neural network of the present invention are illustrated by specific examples below. Figure 2 The network architecture shown in the figure shows the interactive relationship between the redundancy analysis module, path sampling module, and multi-level aggregation module, as well as the data flow direction.

[0059] In this example, a specific heterogeneous graph neural network architecture is used as the base model, trained with the Adam optimizer, with a weight decay rate of 1e-4, a learning rate of 0.005, and a batch size of 16. Furthermore, a meta-path dependency matrix and a meta-path redundancy graph are constructed to optimize meta-path sampling, and a differentiable modular architecture search strategy is employed to identify adaptive meta-path aggregation strategies.

[0060] 1. Verification of the accuracy of the training method The proposed training method is compared with existing hypergraph neural network (HGNN) model training methods and validated on the Digital Library and Literature Project (DBLP) dataset, the Internet Movie Database (IMDB) dataset, and the Association for Computing Machinery (ACM) dataset. Recent heterogeneous graph neural network model training methods include semantically enhanced heterogeneous graph neural network (SeHGNN) and long-range metapath search network (LMSPS) based on progressive sampling.

[0061] Verification results show that the training method of the present invention effectively improves the accuracy of heterogeneous graph neural networks and is superior to existing heterogeneous graph neural network model training methods. In experiments, the present invention achieved significant performance improvements on the Internet Movie Database dataset. Both the macro-average F1 (Macro-F1) and micro-average F1 (Micro-F1) indicators showed accuracy in movie classification and user recommendation tasks, verifying the effectiveness and advantages of the present invention in recommendation system applications. Experiments on the Digital Library and Documentation Project dataset and the Computer Association dataset showed that the present invention performed well in tasks such as author classification and research field classification. The Macro-F1, Micro-F1, and test accuracy indicators were superior to existing methods, fully demonstrating the excellent performance and broad application prospects of the present invention in academic relationship mining tasks.

[0062] 2. Validation of Meta-path Search Algorithm Two mainstream heterogeneous graph neural network models, a semantically enhanced heterogeneous graph neural network and a progressive sampling-based long-range meta-path search network, were selected. The meta-paths found by the proposed method were substituted for the predefined meta-path set of the HGNN model to compare performance changes. Verification results showed that after replacing the meta-paths, the semantically enhanced heterogeneous graph neural network improved by 4.8%, and the long-range meta-path search network based on progressive sampling improved by 3.7%. This fully demonstrates the effectiveness and strong generalization ability of the search results of the proposed method, reflecting its significant advantages in meta-path mining.

[0063] 3. Verification of the key role of the meta-path search stage in overall model performance We thoroughly analyzed the key role of the meta-path search stage in our method on overall model performance. Through ablation experiments, we compared the performance of the complete model with that of a model that removed the meta-path search optimization stage on four datasets: DBLP, IMDB, ACM, and the Open Graph Benchmark-Microsoft Academic Graph (OGBN-MAG).

[0064] Verification results show that the complete model including the meta-path search optimization phase achieved a test accuracy of 56.20%, while the model without this phase achieved only 52.45%. This significant difference highlights the central role of the meta-path search phase in improving model performance, demonstrating its effective reduction of information redundancy and enhanced representation of heterogeneous graph data.

[0065] The above validation results demonstrate that the method provided by this paper exhibits significant advantages in the field of heterogeneous graph neural networks. Its adaptive meta-path search and aggregation strategy not only effectively improves model performance and efficiency, but also verifies the superiority of the search algorithm and the necessity of key stages, providing solid technical support for the practical application of heterogeneous graph neural networks.

[0066] The search model establishment method based on heterogeneous graph neural network provided by the present invention, compared with the existing heterogeneous graph neural network model based on meta-path, optimizes meta-path sampling by constructing a meta-path dependency matrix and a meta-path redundancy graph, thereby avoiding the appearance of redundant information in the model during the sampling process and improving sampling efficiency; when identifying meta-path combinations, only the correlation matrix and graph structure need to be constructed, and a large number of additional parameters will not be introduced, and the data occupies a small space; at the same time, a differentiable modular architecture search strategy is also proposed, which can adaptively aggregate information for different meta-path combinations and downstream tasks, effectively improving the model's feature extraction ability for complex graph structures, and ultimately improving the efficiency and accuracy of the model in heterogeneous graph learning tasks.

[0067] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present invention also provides a search method based on a heterogeneous graph neural network, the method comprising: using the optimized search model in the search model establishment method based on a heterogeneous graph neural network in the above-mentioned embodiment to search for comprehensive feature information related to the target node.

[0068] For example, the target node is user A. By inputting user A's identity information into the optimized search model, information such as orders, categories, products, and other related users associated with user A can be searched.

[0069] Based on the same inventive concept, corresponding to any of the above embodiments and methods, the present invention also provides a search model establishment system based on heterogeneous graph neural networks. Figure 3The search model establishment system 300 based on heterogeneous graph neural network includes a collection module 310, a generation and search module 320, a first optimization module 330, a establishment module 340 and a second optimization module 350.

[0070] The collecting module 310 , for example, executes step S100 to collect meta-paths related to the target node to form an initial meta-path set.

[0071] The generation and search module 320, for example, executes step S200 to generate a meta-path redundancy graph using meta-paths in the initial meta-path set whose pairwise dependencies are higher than a preset threshold; and searches for an optimal non-adjacent meta-path combination in the meta-path redundancy graph as the initial non-redundant meta-path combination.

[0072] The first optimization module 330 , for example, executes step S300 to optimize the architecture parameters and network weights of the initial non-redundant meta-path combination according to the meta-paths and meta-path strengths in the initial non-redundant meta-path combination to obtain an optimal non-redundant meta-path combination.

[0073] The establishing module 340 , for example, executes step S400 to establish a meta-path aggregation search space as a search model for the optimal non-redundant meta-path combination.

[0074] The second optimization module 350, for example, executes step S500 to optimize the training search model using a differentiable modular architecture search strategy.

[0075] Through the collaborative cooperation of the above-mentioned components, the search model establishment system based on heterogeneous graph neural network provided by the embodiment of the present invention uses the meta-paths in the initial meta-path set whose pairwise dependencies are higher than a preset threshold to generate a meta-path redundancy graph, effectively reducing information redundancy and accurately selecting the most appropriate meta-path combination; with the help of a micro-modular architecture search strategy, the structure and semantic information of the heterogeneous graph are comprehensively mined, which significantly improves the performance and applicability of the search model, thereby improving the accuracy and efficiency of the search results.

[0076] The further functional description of each of the above modules is the same as that of the above corresponding method embodiments and will not be repeated here.

[0077] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, it implements the search model establishment method based on a heterogeneous graph neural network and / or the search method based on a heterogeneous graph neural network described in any of the above embodiments.

[0078] Figure 4A more specific hardware structure diagram of an electronic device provided in this embodiment is shown. The device may include: a processor 410, a memory 420, an input / output interface 430, a communication interface 440, and a bus 450. The processor 410, the memory 420, the input / output interface 430, and the communication interface 440 are communicatively connected to each other within the device via the bus 450.

[0079] The processor 410 can be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0080] The memory 420 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage devices, dynamic storage devices, etc. The memory 420 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 420 and is called and executed by the processor 410.

[0081] The input / output interface 430 is used to connect to input / output modules to enable information input and output. The input / output modules can be configured as components within the device (not shown) or externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, and various sensors. Output devices may include a display, speaker, vibrator, indicator light, and the like.

[0082] Communication interface 440 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.).

[0083] The bus 450 comprises a pathway for transmitting information between various components of the device, such as the processor 410 , the memory 420 , the input / output interface 430 , and the communication interface 440 .

[0084] It should be noted that although the above device only shows the processor 410, the memory 420, the input / output interface 430, the communication interface 440, and the bus 450, in a specific implementation, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may only include the components necessary to implement the embodiments of this specification, and does not necessarily include all the components shown in the figure.

[0085] The electronic device of the above embodiment is used to implement the corresponding search model establishment method based on heterogeneous graph neural network and / or search method based on heterogeneous graph neural network in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0086] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present invention also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the search model establishment method based on heterogeneous graph neural networks and / or the search method based on heterogeneous graph neural networks as described in any of the above embodiments.

[0087] The computer-readable media of this embodiment includes permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0088] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the search model establishment method based on heterogeneous graph neural network and / or the search method based on heterogeneous graph neural network as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0089] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention (including the claims) is limited to these examples. Within the scope of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present invention as described above, which are not provided in detail for the sake of simplicity.

[0090] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of the present invention, well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided figures. Furthermore, devices may be shown in block diagram form to avoid obscuring the embodiments of the present invention, and this also takes into account the fact that the implementation details of these block diagram devices are highly dependent on the platform on which the embodiments of the present invention will be implemented (i.e., such details should be fully understood by those skilled in the art). Where specific details (e.g., circuits) are set forth to describe exemplary embodiments of the present invention, it will be apparent to those skilled in the art that the embodiments of the present invention can be implemented without or with variations in these specific details. Accordingly, such descriptions should be regarded as illustrative rather than restrictive.

[0091] Although the present invention has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those skilled in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may utilize the discussed embodiments.

[0092] The embodiments of the present invention are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for establishing a search model based on a heterogeneous graph neural network, characterized in that: include: Collect the meta-paths related to the target node to form an initial meta-path set; Generate a meta-path redundancy graph by using the meta-paths in the initial meta-path set whose dependencies between each other are higher than a preset threshold; Finding an optimal non-adjacent meta-path combination from the meta-path redundancy graph as an initial non-redundant meta-path combination; Optimizing the architecture parameters and network weights of the initial non-redundant meta-path combination according to the meta-paths and meta-path strengths in the initial non-redundant meta-path combination to obtain an optimal non-redundant meta-path combination; Establishing a meta-path aggregation search space as a search model for the optimal non-redundant meta-path combination; The search model is optimized and trained using a differentiable modular architecture search strategy.

2. The method for establishing a search model based on a heterogeneous graph neural network according to claim 1, characterized in that: The dependencies are: ; in, Metapath and metapath The dependence between 、 Metapath , meta path The set of nodes covered, 、 Metapath , meta path The set of covered edges.

3. The method for establishing a search model based on a heterogeneous graph neural network according to claim 1, characterized in that: The meta-path strength is: ; in, Metapath The meta-path strength of 、 and meta-path respectively , meta path The associated schema parameters, is the total number of meta-paths in the initial non-redundant meta-path combination, 、 Metapath , meta path The size of the node set covered, 、 Metapath , meta path The size of the edge set covered, is the temperature parameter.

4. The method for establishing a search model based on a heterogeneous graph neural network according to claim 1, wherein: Optimizing the architecture parameters and network weights of the initial non-redundant meta-path combination specifically includes: alternately updating the architecture parameters and the network weights until optimal architecture parameters and network weights are obtained.

5. The method for establishing a search model based on a heterogeneous graph neural network according to claim 1, wherein: The meta-path aggregation search space can realize one or more of message passing, feature aggregation and meta-path fusion; The message transmission utilizes direct aggregation method, average aggregation method and / or multi-layer perceptron transformation method to transmit node information in the meta-path; The feature aggregation utilizes an average pooling method, a maximum pooling method, and / or an attention mechanism to aggregate the features after message transmission; The meta-path fusion utilizes an average pooling method, a single-head attention mechanism, and / or a multi-head attention mechanism to fuse the features after feature aggregation, and outputs the meta-path fused features.

6. The method for establishing a search model based on a heterogeneous graph neural network according to claim 1, characterized in that: The search model is optimized and trained using a differentiable modular architecture search strategy, specifically including: Dividing the meta-path aggregation process in the search model into multiple modules, each module containing a set of candidate operations; A softmax-weighted hypernetwork is used to convert discrete architecture choices into continuous probabilities, and the weight parameters in the search model and the model weights are jointly optimized via gradient descent.

7. A search method based on heterogeneous graph neural network, characterized in that: include: The optimized search model in the method according to any one of claims 1 to 6 is used to search for comprehensive feature information related to the target node.

8. A search model building system based on heterogeneous graph neural network, characterized in that: include: A collection module is used to collect meta-paths related to the target node to form an initial meta-path set; a generating and searching module, configured to generate a meta-path redundancy graph by using meta-paths in the initial meta-path set whose dependencies between each other are higher than a preset threshold; Finding an optimal non-adjacent meta-path combination from the meta-path redundancy graph as an initial non-redundant meta-path combination; a first optimization module, configured to optimize the architecture parameters and network weights of the initial non-redundant meta-path combination according to the meta-paths and meta-path strengths in the initial non-redundant meta-path combination to obtain an optimal non-redundant meta-path combination; An establishing module, configured to establish a meta-path aggregation search space as a search model for the optimal non-redundant meta-path combination; The second optimization module is used to optimize and train the search model using a differentiable modular architecture search strategy.

9. An electronic 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 program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images