Edge application screening method, terminal device, and medium

Abstract

This invention discloses a method for screening edge applications, a terminal device, and a computer-readable storage medium. The method includes: determining attention weights between the edge application node and its neighboring nodes, wherein the neighboring nodes include industry customer nodes and application scenario nodes; fusing a first feature corresponding to the industry customer node and a second feature corresponding to the edge application node according to the attention weights to obtain a first fused feature; fusing a third feature corresponding to the application scenario node and the second feature according to the attention weights to obtain a second fused feature; determining a target feature corresponding to the edge application node based on the first fused feature and the second fused feature; and determining the screening result of the edge application corresponding to the edge application node based on the target feature. This invention aims to improve the accuracy of the screening results for edge applications.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a screening method, terminal device, and computer-readable storage medium for edge applications. Background Technology

[0002] In the era of 5G mobile communication, the Internet of Things (IoT) is expanding rapidly, and the massive number of IoT devices will create bottlenecks in data processing and cost / energy consumption for cloud computing. At the same time, a superior user experience requires cloud content to be extended to users. Therefore, the rapid development of Mobile Edge Computing (MEC) is an inevitable technological evolution. MEC provides flexible network access capabilities and edge computing services at the edge of the mobile network, reducing network transmission and service delivery latency, improving data security, and providing new momentum for the development of vertical industries.

[0003] Edge computing nodes can be deployed in tiers based on industry customer needs, including core data centers at the prefecture level, important aggregation data centers, ordinary aggregation data centers, and access park data centers. In related technologies, when deploying third-party edge applications on edge computing nodes, the matching and selection of these applications relies heavily on manual experience. This often results in deployed edge applications failing to provide the functionality required by users. In other words, these technologies suffer from inaccurate third-party edge application selection results.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a method for screening edge applications, a terminal device, and a computer-readable storage medium, with the aim of improving the accuracy of the screening results for edge applications.

[0006] To achieve the above objectives, the present invention provides a method for filtering edge applications, the method comprising the following steps:

[0007] Determine the attention weights between edge application nodes and their neighboring nodes, wherein the neighboring nodes include industry customer nodes and application scenario nodes;

[0008] A first fused feature is obtained by fusing the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node according to the attention weight; and a second fused feature is obtained by fusing the third feature corresponding to the application scenario node and the second feature according to the attention weight.

[0009] The target features corresponding to the edge application node are determined based on the first fusion feature and the second fusion feature;

[0010] The filtering results of the edge applications corresponding to the edge application nodes are determined based on the target features.

[0011] Optionally, the step of determining the attention weights between the edge application node and each of its neighboring nodes includes:

[0012] Obtain the neighbor node features and the second feature corresponding to each of the neighbor nodes;

[0013] The importance of each neighbor node to the edge application node is determined based on a preset shared weight matrix, the neighbor node features corresponding to each neighbor node, and the second feature.

[0014] Based on the importance of each neighboring node to the edge application node, the attention weight between the edge application node and each neighboring node is determined.

[0015] Optionally, the step of determining the filtering result of the edge application corresponding to the edge application node based on the target features includes:

[0016] The target features are input into a pre-trained filter, and the filter determines the filtering result of the edge application.

[0017] Optionally, during the training of the filter, the binary cross-entropy binary logarithmic loss function is used as the objective function for training the filter.

[0018] Optionally, before the steps of fusing the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node according to the attention weight to obtain the first fused feature; and fusing the third feature corresponding to the application scenario node and the second feature according to the attention weight to obtain the second fused feature, the method further includes:

[0019] Obtain the first text feature corresponding to the industry customer node, the second text feature corresponding to the edge application node, and the third text feature corresponding to the application scenario node;

[0020] The first feature is determined based on the first text feature, the second feature is determined based on the second text feature, and the third feature is determined based on the third text feature, wherein the first feature, the second feature, and the third feature are vectors of a preset dimension.

[0021] Optionally, the first text feature includes text features corresponding to industry customer needs; the second text feature includes text features corresponding to edge application evaluations; and the third text feature includes text features corresponding to edge application scenario descriptions.

[0022] Optionally, before the step of determining the attention weights between the edge application node and each of its neighboring nodes, the method further includes:

[0023] Construct a heterogeneous graph network, wherein the heterogeneous graph network is centered on the edge application node, and the industry customer node and the application scenario node are the neighbor nodes of the edge application node;

[0024] The step of determining the attention weights between the edge application node and its neighboring nodes includes:

[0025] The attention weights between the edge application node and each of its neighboring nodes are determined based on the heterogeneous graph network.

[0026] In addition, to achieve the above objectives, the present invention also provides a terminal device, the terminal device including a memory, a processor, and a filtering program for edge applications stored in the memory and executable on the processor, wherein the filtering program for edge applications, when executed by the processor, implements the steps of the edge application filtering method as described above.

[0027] Furthermore, to achieve the above objectives, the present invention also provides a terminal device, the terminal device comprising:

[0028] The determination module is used to determine the attention weight between the edge application node and each neighboring node, wherein the neighboring nodes include industry customer nodes and application scenario nodes;

[0029] The first fusion module is used to fuse the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node according to the attention weight to obtain a first fused feature; and to fuse the third feature corresponding to the application scenario node and the second feature according to the attention weight to obtain a second fused feature.

[0030] The second fusion module is used to determine the target features corresponding to the edge application node based on the first fusion feature and the second fusion feature;

[0031] The filtering module is used to determine the filtering results of the edge application corresponding to the edge application node based on the target features.

[0032] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing an edge application filtering program, which, when executed by a processor, implements the steps of the edge application filtering method as described above.

[0033] This invention proposes a method, terminal device, and computer-readable storage medium for filtering edge applications. First, it determines the attention weights between the edge application node and its neighboring nodes, where the neighboring nodes include industry customer nodes and application scenario nodes. Then, it fuses a first feature corresponding to the industry customer node and a second feature corresponding to the edge application node based on the attention weights to obtain a first fused feature. Next, it fuses a third feature corresponding to the application scenario node and the second feature based on the attention weights to obtain a second fused feature. Finally, it determines the target feature corresponding to the edge application node based on the first and second fused features. Finally, it determines the filtering result of the edge application corresponding to the edge application node based on the target feature. Because it can comprehensively filter edge applications based on multiple factors such as industry customers and edge application scenarios, it improves the accuracy of the edge application filtering results. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the terminal structure of the hardware operating environment involved in the embodiments of the present invention;

[0035] Figure 2 This is a flowchart illustrating an embodiment of the edge application filtering method of the present invention;

[0036] Figure 3 This is a schematic diagram of a heterogeneous graph network according to an embodiment of the present invention;

[0037] Figure 4 This is a flowchart illustrating an optional implementation scheme according to an embodiment of the present invention;

[0038] Figure 5 This is a schematic diagram of the screening process involved in an embodiment of the present invention;

[0039] Figure 6 This is a modular schematic diagram of the terminal device involved in an embodiment of the present invention.

[0040] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0041] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0042] like Figure 1 As shown, Figure 1 This is a schematic diagram of the terminal structure of the hardware operating environment involved in the embodiments of the present invention.

[0043] like Figure 1As shown, the control terminal may include: a processor 1001, such as a CPU, a network interface 1003, a memory 1004, and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The network interface 1003 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1004 may be high-speed RAM or stable non-volatile memory, such as a disk drive. Alternatively, the memory 1004 may be a storage device independent of the aforementioned processor 1001.

[0044] Those skilled in the art will understand that Figure 1 The terminal structure shown does not constitute a limitation on the terminal and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0045] like Figure 1 As shown, the memory 1004, which serves as a computer storage medium, may include an operating system, a network communication module, and a screening program for edge applications.

[0046] exist Figure 1 In the terminal shown, the processor 1001 can be used to call the edge application filtering program stored in the memory 1004 and perform the following operations:

[0047] Determine the attention weights between edge application nodes and their neighboring nodes, wherein the neighboring nodes include industry customer nodes and application scenario nodes;

[0048] A first fused feature is obtained by fusing the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node according to the attention weight; and a second fused feature is obtained by fusing the third feature corresponding to the application scenario node and the second feature according to the attention weight.

[0049] The target features corresponding to the edge application node are determined based on the first fusion feature and the second fusion feature;

[0050] The filtering results of the edge applications corresponding to the edge application nodes are determined based on the target features.

[0051] Furthermore, the processor 1001 can invoke the edge application filtering program stored in the memory 1004 and also perform the following operations:

[0052] Obtain the neighbor node features and the second feature corresponding to each of the neighbor nodes;

[0053] The importance of each neighbor node to the edge application node is determined based on a preset shared weight matrix, the neighbor node features corresponding to each neighbor node, and the second feature.

[0054] Based on the importance of each neighboring node to the edge application node, the attention weight between the edge application node and each neighboring node is determined.

[0055] Furthermore, the processor 1001 can invoke the edge application filtering program stored in the memory 1004 and also perform the following operations:

[0056] The target features are input into a pre-trained filter, and the filter determines the filtering result of the edge application.

[0057] Furthermore, the processor 1001 can invoke the edge application filtering program stored in the memory 1004 and also perform the following operations:

[0058] Obtain the first text feature corresponding to the industry customer node, the second text feature corresponding to the edge application node, and the third text feature corresponding to the application scenario node;

[0059] The first feature is determined based on the first text feature, the second feature is determined based on the second text feature, and the third feature is determined based on the third text feature, wherein the first feature, the second feature, and the third feature are vectors of a preset dimension.

[0060] Furthermore, the processor 1001 can invoke the edge application filtering program stored in the memory 1004 and also perform the following operations:

[0061] Construct a heterogeneous graph network, wherein the heterogeneous graph network is centered on the edge application node, and the industry customer node and the application scenario node are the neighbor nodes of the edge application node;

[0062] The step of determining the attention weights between the edge application node and its neighboring nodes includes:

[0063] The attention weights between the edge application node and each of its neighboring nodes are determined based on the heterogeneous graph network.

[0064] In the 5G era, the Internet of Things (IoT) is expanding rapidly, and the massive number of IoT devices will create bottlenecks in data processing and energy consumption for cloud computing. Simultaneously, an exceptional user experience requires cloud content to be extended to users. Therefore, the rapid development of MEC (Multi-access Edge Computing) is an inevitable technological evolution. MEC reduces network transmission and service delivery latency by providing flexible network access capabilities and edge computing services at the mobile network edge, thereby improving data security and providing new impetus for the development of vertical industries.

[0065] Edge computing nodes can be deployed in tiers according to industry customer needs, including core data centers at the prefecture level, important aggregation data centers, ordinary aggregation data centers, and access park data centers. In actual deployment, a unified cloud-based infrastructure integrating CT / IT (Communication Technology / Information Technology) can be built to support UPF (User Port Function), MEC, and proprietary or carefully selected third-party edge applications. These edge applications can include video rendering, image processing, text processing, computing power scheduling, and hardware acceleration.

[0066] Because different industry users have varying functional requirements for edge applications, edge applications typically need to be screened to ensure they match user needs. This screening process determines the most suitable edge application for deployment on the appropriate edge computing nodes. However, in related technologies, third-party edge application matching and screening largely relies on human experience, often resulting in deployed edge applications failing to provide the functionality required by users. In other words, these technologies suffer from the flaw of inaccurate third-party edge application screening results.

[0067] To address the aforementioned shortcomings of related technologies, this invention proposes a method for screening edge applications. Leveraging the ability of graph attention networks to assign different weights based on the varying influence of neighboring nodes, a multi-level heterogeneous graph attention network is proposed. This network constructs a heterogeneous graph network centered on the edge application, composed of three types of heterogeneous nodes: industry customers, edge applications, and edge application scenarios. A node-level attention mechanism determines the varying importance of neighboring nodes (industry customer nodes and edge application scenario nodes) of an edge application node in determining whether the edge application is abnormal, and assigns different weights to the relationships between nodes based on their importance. This achieves the goal of comprehensively screening edge applications based on multiple factors, including industry customers and edge application scenarios, thereby improving the accuracy of edge application screening results.

[0068] The following specific embodiments further explain the edge application screening method proposed in this invention.

[0069] In one embodiment, please refer to Figure 2 The edge application filtering method includes the following steps:

[0070] Step S10: Determine the attention weights between the edge application node and each neighboring node, wherein the neighboring nodes include industry customer nodes and application scenario nodes;

[0071] Step S20: Based on the attention weight, fuse the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node to obtain a first fused feature; and based on the attention weight, fuse the third feature corresponding to the application scenario node and the second feature to obtain a second fused feature;

[0072] Step S30: Determine the target features corresponding to the edge application node based on the first fusion feature and the second fusion feature;

[0073] Step S40: Determine the filtering results of the edge application corresponding to the edge application node based on the target features.

[0074] In this embodiment, the system first collects a set of industry customer business requirements, an edge application evaluation set, and an edge application scenario description set from the edge computing management module (MEPM, Mobile Edge PlatformManager) within the most recent time period T as basic data. Then, a heterogeneous network graph is constructed based on this basic data. Finally, the attention weights between the edge application node and its neighboring nodes are determined based on this heterogeneous network graph.

[0075] For example, please refer to Figure 3 The heterogeneous network graph contains three types of nodes: industry customer nodes, edge application nodes, and application scenario nodes. The heterogeneous network graph is centered on the edge application node, with the industry customer node and the application scenario node serving as its neighbors.

[0076] It is understandable that, since edge applications provided by operators are generally those corresponding to essential functions, the nodes corresponding to edge applications in the heterogeneous network diagram presented in this embodiment are generally the nodes corresponding to third-party edge applications. Furthermore, customer nodes or application scenario nodes within the same industry can be associated with one or more edge application nodes.

[0077] It should be noted that in the heterogeneous network graph, the characteristics of industry customer nodes are determined based on the text features corresponding to industry customer needs. That is, the set of industry customer business needs containing the business needs of various industry customers is used as the characteristic of the industry customer node, which can be represented as B = {b1, b2, b3, ...}. The characteristics of edge application nodes are determined based on the text features corresponding to edge application evaluations. That is, the set of edge application evaluations containing the main usage evaluation descriptions of each edge application is used as the characteristic of the edge application node, which can be represented as P = {p1, p2, p3, ...}. The characteristics of application scenario nodes are determined based on the text features corresponding to edge application scenario descriptions. That is, the set of edge application scenario descriptions with corresponding text features is used as the characteristic of the application scenario node, which can be represented as U = {u1, u2, u3, ...}. It is understandable that when constructing the heterogeneous network graph, after obtaining the basic data, the basic data can be preprocessed. This involves serializing the industry customer business need set, edge application evaluation set, and edge application scenario description set into text and encoding the feature text into a sequence representation. Then, the heterogeneous network graph is constructed based on the preprocessed basic data.

[0078] Furthermore, once the heterogeneous network graph is determined, the attention weights between each edge application node and its associated neighboring nodes can be determined based on the heterogeneous network graph.

[0079] For example, the neighbor node features and the second feature corresponding to each neighbor node can be obtained first. Then, the importance of each neighbor node to the edge application node can be determined based on a preset shared weight matrix, the neighbor node features and the second feature corresponding to each neighbor node. And the attention weight between the edge application node and each neighbor node can be determined according to the importance of each neighbor node to the edge application node.

[0080] In this example, as an optional implementation, since the features corresponding to each node are text features represented by a sequence, word embedding can be used to convert each word into a vector before computation. The input data dimension is z, and the output is set to convert the words into N-dimensional (e.g., 64-dimensional) spatial vectors. The input sequence length is F, so the shape of the output data of this layer is (None, F, N). This achieves vector mapping of the input words, converting the index of each word into an N-dimensional fixed-shape vector. Optionally, when determining the attention weights between edge application nodes and neighboring nodes, the number of convolutional kernels in the graph attention layer can be set to 256, and the activation function can be set to "ReLU". It is understood that the first feature corresponding to the industry customer node, the second feature corresponding to the edge application node, and the third feature corresponding to the application scenario node can be a converted vector or a text feature before conversion. When it is a text feature, the text feature is mapped to a vector before the following operations; otherwise, the attention weights are determined directly based on the following method.

[0081] Furthermore, after obtaining the neighbor node features corresponding to each of the neighbor nodes and the second feature, the attention weight between the edge application node and each neighbor node can be determined. For example, after obtaining the second feature corresponding to the edge application node and the first feature corresponding to a specific industry customer node, the attention coefficient between the edge application node and the industry customer node can be calculated first, and then the attention weight between them can be obtained by weighted summation. Then, based on the attention weight, the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node are fused according to the attention weight to obtain a first fused feature; and the third feature corresponding to the application scenario node and the second feature are fused according to the attention weight to obtain a second fused feature.

[0082] For example, the currently determined second feature is, [characteristic]. The first fusion feature or the second fusion feature is

[0083] Furthermore, an F×F' shared weight matrix W can be applied to each node, and then the attention coefficients can be calculated. These coefficients can represent the importance of each industry customer node j relative to the edge application node i.

[0084]

[0085] Furthermore, to make the attention coefficients easier to calculate and compare, softmax is introduced to regularize all industry customer nodes j adjacent to edge application node i:

[0086]

[0087] Where, α ij Ni represents the attention weight between node pair (i,j), and Ni represents the set of neighbors of node i, i.e., the set of industry customer nodes.

[0088] Furthermore, the first fusion feature can be determined as:

[0089] It should be noted that the second fusion feature can also be determined using the above method.

[0090] Furthermore, after determining the first fusion feature and the second fusion feature, the first fusion feature and the second fusion feature can be applied to the features of the target neighbor nodes of the edge application node. Then, based on the above method, the first fusion feature, the second fusion feature, and the second feature are fused to obtain the target feature corresponding to the edge application node. The filtering result of the edge application corresponding to the edge application node is then determined based on the target feature.

[0091] Alternatively, as one implementation method, please refer to Figure 4 Step S40 includes:

[0092] Step S41: Input the target features into a pre-trained filter, and determine the filtering result of the edge application through the filter.

[0093] In this implementation scheme, the filter can be trained based on pre-prepared training data. The training data can be collected from historical edge application-related datasets in the edge computing management module. A heterogeneous graph network of edge applications is constructed, centered on the application edge and composed of three types of heterogeneous nodes: industry customers, edge applications, and edge application scenarios. The dataset consists of text features of the three types of heterogeneous nodes. The three types of heterogeneous data are processed into text serialization, and the filtering results for each third-party edge application are manually labeled.

[0094] The industry customer business requirements set, edge application evaluation set, and edge application scenario description set are encoded into sequence representations respectively. The length of the encoded sequence for each node feature is defined as F, and the longest length F in the dataset is taken as its encoded sequence length. The length of each data item is padded to F, and its dictionary size is z.

[0095] Manual labeling: The selection results for each third-party edge application in each heterogeneous topology graph are manually labeled, with an N*M shape (representing the M existing third-party edge applications). Finally, the total dataset is divided into training data and test data. For example, 80% of the entire dataset can be used as training data, and the remaining 20% ​​as test data. The training set is used for training, and the test set is used to evaluate and validate the model.

[0096] Furthermore, the filter consists of a fully connected (Dense) layer: the number of neurons is set to M (representing the existing M edge applications), the activation function is set to "sigmoid", and the output 1 represents that the third-party edge application is selected, and 0 represents that the third-party edge application is not selected.

[0097] Then, the error between the predicted edge application screening result and the correct third-party edge application screening result is calculated, and the training objective is to minimize this error. The objective function chosen is the binary cross-entropy loss function. Optionally, the number of training epochs can be set to 1000 (epochs = 1000), and the Adam optimizer is chosen to improve the learning speed of traditional gradient descent (optimizer = 'adam'). The neural network finds the optimal weight values ​​that minimize the objective function through gradient descent, and the neural network learns these weight values ​​autonomously through training. Training is performed using a training set to minimize the objective function, and the model is evaluated and validated using a test set after each training epoch. The weights of the model are derived after convergence.

[0098] After the filter training is completed, the target features corresponding to the edge applications to be filtered can be determined based on the above steps S10 to S40. Then, the filter can be used to determine the filtering result of the edge application based on the target features.

[0099] For example, in a specific implementation scheme, please refer to Figure 5 The edge application filtering process includes:

[0100] Branch 1: Learn the weights of the importance of the first-class neighbor nodes "industry customers" of edge application node i. Input the features h of the third-party edge application node i. bi and the characteristics of its industry customer nodes h pi After being transformed into vectors through word embedding layers, the vectors are input into the neighbor node-level graph attention layer to obtain the first fused feature h, which incorporates the "industry customer" feature information of the neighbor nodes. bi1 .

[0101] Branch 2: Learn the weights of the importance of the second-class neighbor nodes "application scenario" of edge application node i. Input the features h of edge application node i. biand application scenario node characteristics h ui After being transformed into vectors through word embedding layers, the vectors are input into the neighbor node-level graph attention layer to obtain the second fused feature h, which incorporates the "application scenario" feature information of the neighbor nodes. bi2 .

[0102] The word embedding layer converts each word into a vector using word embedding. The input data dimension is z, and the output is set to convert the words into 64-dimensional spatial vectors. The input sequence length is F, so the shape of the output data of this layer is (None, F, 64). The neighbor node-level graph attention layer has 256 convolutional kernels and the activation function is set to "ReLU".

[0103] Furthermore, the first fused feature h, which combines the industry customer characteristics from the outputs of the above two branches, is then... bi1 And the second fusion feature h after incorporating application scenario features. bi2 The input is fed into the central node-level graph attention layer for central node-level attention aggregation, and the output is the new feature of the edge application node that simultaneously fuses the features of two types of neighboring nodes, namely the target feature h. bi Finally, the target feature h bi The input is fed into a filter consisting of fully connected layers. The filter determines the filtering result applied to this edge. The central node-level graph attention layer has 128 convolutional kernels and the activation function is set to "relu".

[0104] In the technical solution disclosed in this embodiment, the attention weights between the edge application node and each neighboring node are first determined, wherein the neighboring nodes include industry customer nodes and application scenario nodes. Then, a first fused feature is obtained by fusing a first feature corresponding to the industry customer node and a second feature corresponding to the edge application node according to the attention weights; and a second fused feature is obtained by fusing a third feature corresponding to the application scenario node and the second feature according to the attention weights. A target feature corresponding to the edge application node is then determined based on the first fused feature and the second fused feature. Finally, the screening result of the edge application corresponding to the edge application node is determined based on the target feature. Since edge applications can be screened comprehensively based on multiple factors including industry customers and edge application scenarios, the accuracy of the edge application screening results is improved.

[0105] Furthermore, this embodiment of the invention also proposes a terminal device, the terminal device including: a memory, a processor, and a filtering program for edge applications stored in the memory and executable on the processor, wherein when the edge application filtering program is executed by the processor, it implements the steps of the edge application filtering method as described in the above embodiments.

[0106] In addition, please refer to Figure 6 The present invention also proposes a terminal device 100, the terminal device 100 comprising:

[0107] The determining module 101 is used to determine the attention weight between the edge application node and each neighboring node, wherein the neighboring nodes include industry customer nodes and application scenario nodes;

[0108] The first fusion module 102 is used to fuse the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node according to the attention weight to obtain a first fused feature; and to fuse the third feature corresponding to the application scenario node and the second feature according to the attention weight to obtain a second fused feature.

[0109] The second fusion module 103 is used to determine the target features corresponding to the edge application node based on the first fusion feature and the second fusion feature;

[0110] The filtering module 104 is used to determine the filtering result of the edge application corresponding to the edge application node based on the target features.

[0111] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a filtering program for edge applications, wherein when the filtering program for edge applications is executed by a processor, it implements the steps of the edge application filtering method as described in the above embodiments.

[0112] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0113] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0114] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (data processing device or server) to execute the methods described in the various embodiments of the present invention.

[0115] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for filtering edge applications, characterized in that, The filtering method for edge applications includes: Determine the attention weights between edge application nodes and their neighboring nodes, wherein the neighboring nodes include industry customer nodes and application scenario nodes; A first fused feature is obtained by fusing the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node according to the attention weight; and a second fused feature is obtained by fusing the third feature corresponding to the application scenario node and the second feature according to the attention weight; wherein, the first feature is determined based on the text features corresponding to industry customer needs, the second feature is determined based on the text features corresponding to edge application evaluation, and the third feature is determined based on the text features corresponding to edge application scenario description. The target features corresponding to the edge application node are determined based on the first fusion feature and the second fusion feature; The filtering results of the edge applications corresponding to the edge application nodes are determined based on the target features.

2. The edge application filtering method according to claim 1, characterized in that, The step of determining the attention weights between the edge application node and its neighboring nodes includes: Obtain the neighbor node features and the second feature corresponding to each of the neighbor nodes; The importance of each neighbor node to the edge application node is determined based on a preset shared weight matrix, the neighbor node features corresponding to each neighbor node, and the second feature. Based on the importance of each neighboring node to the edge application node, the attention weight between the edge application node and each neighboring node is determined.

3. The edge application filtering method according to claim 1, characterized in that, The step of determining the filtering results of the edge application corresponding to the edge application node based on the target features includes: The target features are input into a pre-trained filter, and the filter determines the filtering result of the edge application.

4. The edge application filtering method according to claim 3, characterized in that, During the training process of the filter, the binary cross-entropy binary logarithmic loss function is used as the objective function for training the filter.

5. The edge application filtering method according to claim 1, characterized in that, The first fused feature is obtained by fusing the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node according to the attention weight; Before the step of fusing the third feature corresponding to the application scenario node and the second feature according to the attention weight to obtain the second fused feature, the method further includes: Obtain the first text feature corresponding to the industry customer node, the second text feature corresponding to the edge application node, and the third text feature corresponding to the application scenario node; wherein, the first text feature includes the text feature corresponding to the industry customer demand, the second text feature includes the text feature corresponding to the edge application evaluation, and the third text feature includes the text feature corresponding to the edge application scenario description. The first feature is determined based on the first text feature, the second feature is determined based on the second text feature, and the third feature is determined based on the third text feature, wherein the first feature, the second feature, and the third feature are vectors of a preset dimension.

6. The edge application filtering method according to claim 5, characterized in that, Before the step of determining the attention weights between the edge application node and each of its neighboring nodes, the method further includes: Construct a heterogeneous graph network, wherein the heterogeneous graph network is centered on the edge application node, and the industry customer node and the application scenario node are the neighbor nodes of the edge application node; The step of determining the attention weights between the edge application node and its neighboring nodes includes: The attention weights between the edge application node and each of its neighboring nodes are determined based on the heterogeneous graph network.

7. A terminal device, characterized in that, The terminal device includes: a memory, a processor, and a filtering program for edge applications stored in the memory and executable on the processor. When the edge application filtering program is executed by the processor, it implements the steps of the edge application filtering method as described in any one of claims 1 to 6.

8. A terminal device, characterized in that, The terminal device includes: The determination module is used to determine the attention weight between the edge application node and each neighboring node, wherein the neighboring nodes include industry customer nodes and application scenario nodes; The first fusion module is used to fuse the first feature corresponding to the industry customer node and the second feature corresponding to the edge application node according to the attention weight to obtain a first fused feature; and to fuse the third feature corresponding to the application scenario node and the second feature according to the attention weight to obtain a second fused feature; wherein, the first feature is determined according to the text feature corresponding to the industry customer demand, the second feature is determined according to the text feature corresponding to the edge application evaluation, and the third feature is determined according to the text feature corresponding to the edge application scenario description. The second fusion module is used to determine the target features corresponding to the edge application node based on the first fusion feature and the second fusion feature; The filtering module is used to determine the filtering results of the edge application corresponding to the edge application node based on the target features.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a filtering program for edge applications, which, when executed by a processor, implements the steps of the edge application filtering method as described in any one of claims 1 to 6.

Images