CN115705414A - Screening method of edge application, terminal device and medium - Google Patents

Abstract

The invention discloses a screening method of edge application, terminal equipment and a computer readable storage medium. The method comprises the following steps: determining attention weights between the edge application nodes and adjacent nodes, wherein the adjacent nodes comprise industry customer nodes and application scene 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 weight to obtain a first fused feature; fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature; determining a target feature corresponding to the edge application node according to the first fusion feature and the second fusion feature; and determining a screening result of the edge application corresponding to the edge application node according to the target characteristics. The invention aims to achieve the effect of improving the accuracy of the screening result of the edge application.

Description

Screening method of edge application, terminal device and medium

Technical Field

The present invention relates to the field of data processing technologies, and in particular, to a method for screening edge applications, a terminal device, and a computer-readable storage medium.

Background

In the fifth generation of mobile communication, 5G, all things are interconnected, mass things are upwards extended, and bottlenecks are generated in the aspects of cloud computing mode data processing and cost and energy consumption. Meanwhile, the extreme user experience also needs the cloud content to extend to the user, and therefore, the rapid development of multi-access Edge Computing (MEC) will be a necessity of technical evolution. The MEC provides flexible network access capability and edge computing service at the edge of a mobile network, reduces network transmission and service delivery time delay, improves data security, and provides new development kinetic energy for the vertical industry.

The edge computing nodes can be hierarchically deployed in a core computer room, an important convergence computer room, a common convergence computer room and an access park computer room of a city according to the requirements of industry customers. In the related art, when a third-party edge application is deployed in an edge computing node, matching and screening of the third-party edge application mainly depends on manual experience, so that a phenomenon that the deployed edge application cannot provide a function matched with the requirement of a user often occurs. Namely, the related art has the defect that the screening result of the third-party edge application is inaccurate.

The above is only for the purpose of assisting understanding of the technical aspects of the present invention, and does not represent an admission that the above is prior art.

Disclosure of Invention

The invention mainly aims to provide a screening method of edge application, terminal equipment and a computer readable storage medium, aiming at achieving the effect of improving the accuracy of the screening result of the edge application.

In order to achieve the above object, the present invention provides a screening method for edge application, including the steps of:

determining attention weights between the edge application nodes and each adjacent node, wherein the adjacent nodes comprise industry client nodes and application scene nodes;

fusing a first feature corresponding to the industry client node and a second feature corresponding to the edge application node according to the attention weight to obtain a first fused feature; fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature;

determining a target feature corresponding to the edge application node according to the first fusion feature and the second fusion feature;

and determining a screening result of the edge application corresponding to the edge application node according to the target characteristics.

Optionally, the step of determining attention weights between the edge application node and each neighboring node includes:

acquiring adjacent node characteristics and the second characteristics corresponding to each adjacent node;

determining the importance of each neighbor node to the edge application node based on a preset shared weight matrix, the neighbor node characteristics corresponding to each neighbor node and the second characteristics;

and determining attention weights between the edge application node and each adjacent node according to the importance of each adjacent node to the edge application node.

Optionally, the step of determining a screening result of the edge application corresponding to the edge application node according to the target feature includes:

inputting the target features into a pre-trained filter, and determining a filtering result of the edge application through the filter.

Optionally, in the training process of the filter, a binary cross entropy binary logarithmic loss function is used as an objective function for training the filter.

Optionally, the first feature corresponding to the industry client node and the second feature corresponding to the edge application node are fused according to the attention weight to obtain a first fused feature; before the step of fusing the third feature and the second feature corresponding to the application scene node according to the attention weight to obtain a second fused feature, the method further includes:

acquiring a first text feature corresponding to the industry client node, a second text feature corresponding to the edge application node and a third text feature corresponding to the application scene node;

determining the first feature according to the first text feature, determining the second feature according to the second text feature, and determining the third feature according to the third text feature, wherein the first feature, the second feature, and the third feature are vectors of preset dimensions.

Optionally, the first text feature comprises a text feature corresponding to an industry customer requirement; the second text features comprise text features corresponding to the edge application evaluation; the third text feature comprises a text feature corresponding to the edge application scene description.

Optionally, before the step of determining the attention weight between the edge application node and each neighboring node, the method further includes:

constructing a heteromorphic graph network, wherein the heteromorphic graph network takes the edge application node as a center, and takes the industry customer node and the application scene node as adjacent nodes of the edge application node;

the step of determining attention weights between the edge application node and each neighboring node comprises:

determining attention weights between the edge application node and respective neighboring nodes based on the heterogeneous graph network.

In addition, to achieve the above object, the present invention further provides a terminal device, where the terminal device includes a memory, a processor, and a filter of an edge application stored on the memory and executable on the processor, and the filter of the edge application, when executed by the processor, implements the steps of the method for filtering an edge application as described above.

In addition, to achieve the above object, the present invention further provides a terminal device, including:

the determining module is used for determining attention weights between the edge application nodes and adjacent nodes, wherein the adjacent nodes comprise industry client nodes and application scene nodes;

the first fusion module is used for fusing a first feature corresponding to the industry client node and a second feature corresponding to the edge application node according to the attention weight to obtain a first fusion feature; fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature;

the second fusion module is used for determining a target feature corresponding to the edge application node according to the first fusion feature and the second fusion feature;

and the screening module is used for determining the screening result of the edge application corresponding to the edge application node according to the target characteristics.

In addition, to achieve the above object, the present invention also provides a computer readable storage medium, on which a filter program of an edge application is stored, which when executed by a processor implements the steps of the filter method of the edge application as described above.

According to the screening method, the terminal device and the computer readable storage medium for the edge application, attention weights between edge application nodes and adjacent nodes are determined firstly, wherein the adjacent nodes comprise industry client nodes and application scene nodes, and then first characteristics corresponding to the industry client nodes and second characteristics corresponding to the edge application nodes are fused according to the attention weights to obtain first fusion characteristics; and fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature, determining a target feature corresponding to the edge application node according to the first fused feature and the second fused feature, and finally determining a screening result of the edge application corresponding to the edge application node according to the target feature. Due to the fact that the edge application can be comprehensively screened on the basis of multiple factors of industry customers and edge application scenes, the effect of improving the accuracy of the edge application screening result is achieved.

Drawings

Fig. 1 is a schematic diagram of a terminal structure of a hardware operating environment according to an embodiment of the present invention;

FIG. 2 is a schematic flow chart diagram illustrating an embodiment of a screening method for edge application according to the present invention;

FIG. 3 is a schematic diagram of a heterogeneous graph network according to an embodiment of the present invention;

FIG. 4 is a schematic flow chart of an alternative embodiment in accordance with an example of the present invention;

FIG. 5 is a schematic diagram of a screening process according to an embodiment of the present invention;

fig. 6 is a schematic block diagram of a terminal device according to an embodiment of the present invention.

The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.

Detailed Description

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

As shown in fig. 1, fig. 1 is a schematic terminal structure diagram of a hardware operating environment according to an embodiment of the present invention.

As shown in fig. 1, the control terminal may include: a processor 1001, such as a CPU, a network interface 1003, a memory 1004, a communication bus 1002. Wherein a communication bus 1002 is used to enable connective communication between these components. The network interface 1003 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1004 may be a high-speed RAM memory or a non-volatile memory (e.g., a disk memory). The memory 1004 may alternatively be a storage device separate from the processor 1001.

Those skilled in the art will appreciate that the terminal structure shown in fig. 1 is not intended to be limiting and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.

As shown in fig. 1, the memory 1004, which is a type of computer storage medium, may include therein an operating system, a network communication module, and a filter of an edge application.

In the terminal shown in fig. 1, the processor 1001 may be configured to call a filter of an edge application stored in the memory 1004 and perform the following operations:

determining attention weights between the edge application nodes and each adjacent node, wherein the adjacent nodes comprise industry client nodes and application scene 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 weight to obtain a first fused feature; fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature;

determining a target feature corresponding to the edge application node according to the first fusion feature and the second fusion feature;

and determining a screening result of the edge application corresponding to the edge application node according to the target characteristics.

Further, the processor 1001 may call the filter of the edge application stored in the memory 1004, and further perform the following operations:

acquiring adjacent node characteristics and the second characteristics corresponding to each adjacent node;

determining the importance of each neighbor node to the edge application node based on a preset shared weight matrix, the neighbor node characteristics corresponding to each neighbor node and the second characteristics;

and determining attention weights between the edge application node and each adjacent node according to the importance of each adjacent node to the edge application node.

Further, the processor 1001 may call the filter of the edge application stored in the memory 1004, and further perform the following operations:

inputting the target features into a pre-trained filter, and determining a filtering result of the edge application through the filter.

Further, the processor 1001 may call the filter of the edge application stored in the memory 1004, and further perform the following operations:

acquiring a first text feature corresponding to the industry client node, a second text feature corresponding to the edge application node and a third text feature corresponding to the application scene node;

determining the first feature according to the first text feature, determining the second feature according to the second text feature, and determining the third feature according to the third text feature, wherein the first feature, the second feature, and the third feature are vectors of preset dimensions.

Further, the processor 1001 may call the filter of the edge application stored in the memory 1004, and further perform the following operations:

constructing a heteromorphic graph network, wherein the heteromorphic graph network takes the edge application node as a center, and takes the industry customer node and the application scene node as adjacent nodes of the edge application node;

the step of determining attention weights between the edge application node and respective neighboring nodes comprises:

determining attention weights between the edge application node and respective neighboring nodes based on the heterogeneous graph network.

The 5G time universal object interconnection, the mass internet of things equipment upwards extends, and bottlenecks are generated in the aspects of data processing and cost and energy consumption in a cloud computing mode. At the same time, the ultimate user experience also requires that the content of the cloud be extended to the user. For this reason, rapid development of MECs will be a necessity of technological evolution. The MEC reduces network transmission and service delivery time delay by providing flexible network access capability and edge computing service at the edge of a mobile network, thereby achieving the purposes of improving data security and providing new development kinetic energy for the vertical industry.

The edge computing nodes can be hierarchically deployed in a core computer room, an important convergence computer room, a common convergence computer room and an access park computer room of a city according to the requirements of industry customers. In actual deployment, the cloud base can be constructed to uniformly carry UPF (User Port Function), MEC and edge applications of owned or selected third parties by constructing CT/IT (Communication Technology, communication Technology/Information Technology) fusion. Among other things, edge applications may include video rendering, image processing, text processing, computational scheduling, hardware acceleration, and so forth.

Because different industry users have different functional requirements on the edge application, the edge application generally needs to be screened first in order that the deployed edge application can be matched with the requirements of the users. And determining the most suitable edge application to be deployed to the corresponding edge computing node according to the screening result. However, in the related art, matching screening of the edge application of the third party mainly depends on manual experience, so that a phenomenon that the deployed edge application cannot provide the function matched with the requirement for the user often occurs. Namely, the related art has the defect that the screening result of the third-party edge application is inaccurate.

In order to solve the above-mentioned defects in the related art, an embodiment of the present invention provides a method for screening an edge application, in which a graph attention network is used to assign different weights according to differences in influence of neighboring nodes in the graph network, so as to provide a multi-level heterogeneous graph attention network, and construct an edge application heterogeneous graph network that is centered on an edge application and is composed of three heterogeneous nodes, namely an industry client, an edge application, and an edge application scene. And determining different importance of neighbor nodes of the edge application node, namely an industry client node and an edge application scene node, for judging whether the edge application is abnormal or not through a node level attention mechanism, and distributing different weights for the relationship among the nodes according to the difference of the importance. The purpose of comprehensively screening the edge application based on multiple factors of industry customers and edge application scenes is achieved. Therefore, the effect of improving the accuracy of the edge application screening result is achieved.

The screening method for edge application proposed by the present invention is further explained by specific examples below.

In an embodiment, referring to fig. 2, the method for screening edge applications includes the following steps:

s10, determining attention weights between the edge application nodes and adjacent nodes, wherein the adjacent nodes comprise industry customer nodes and application scene nodes;

s20, fusing a first feature corresponding to the industry client node and a second feature corresponding to the edge application node according to the attention weight to obtain a first fused feature; fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature;

s30, determining a target feature corresponding to the edge application node according to the first fusion feature and the second fusion feature;

and S40, determining a screening result of the edge application corresponding to the edge application node according to the target characteristics.

In this embodiment, an industry client business requirement set, an Edge application evaluation set, and an Edge application scene description set in a recent T period may be collected from an Edge computing management module (MEPM) as basic data. A heterogeneous network graph is then constructed based on the base data. Attention weights between the edge application node and each neighboring node are then determined based on the heterogeneous graph network.

Illustratively, referring to fig. 3, the heterogeneous network graph includes 3 types of nodes, where the 3 types of nodes include an industry customer type node, an edge application type node, and an application scenario type node. The edge application node is used as the center of the network graph of the abnormal graph, and the industry customer node and the application scene node are used as the adjacent nodes of the edge application node.

It can be understood that, in the edge application, the edge application provided by the operator is generally the edge application corresponding to the necessary function. Therefore, in the heterogeneous network diagram provided in this embodiment, the node corresponding to the edge application is generally a node corresponding to a third-party edge application. In addition, the same industry client node or application scene node can have an association relationship with one or more edge application nodes.

It should be noted that, in the heterogeneous network diagram, the characteristics of the industry client node are determined according to the text characteristics corresponding to the industry client requirements, that is, an industry client service requirement set including the service requirements of each industry client is used as the characteristics of the industry client node, and may be represented as B = { B1, B2, B3, … }. The feature of the edge application node is determined according to the text feature corresponding to the edge application evaluation, that is, an edge application evaluation set including the main usage evaluation description of each edge application can be represented as P = { P1, P2, P3, … }. The feature of the application scenario node is determined according to the text feature corresponding to the edge application scenario description, that is, an edge application scenario description set of the text feature corresponding to the edge application scenario description is used as the feature of the application scenario node, and may be represented as U = { U1, U2, U3, … }. It is to be understood that, when constructing the heterogeneous network graph, the basic data may be preprocessed after being obtained. Namely, the industry customer service requirement set, the edge application evaluation set and the edge application scene description set are subjected to text serialization processing, and the feature text is coded into sequence representation. And then constructing the heterogeneous network graph based on the preprocessed basic data.

Further, when determining the heterogeneous network graph, attention weights between each edge application node and its associated neighbor nodes may be determined based on the heterogeneous network graph.

For example, the neighboring node feature and the second feature corresponding to each neighboring node may be obtained first, then the importance of each neighboring node to the edge application node is determined based on a preset shared weight matrix, the neighboring node feature corresponding to each neighboring node, and the second feature, and the attention weight between the edge application node and each neighboring node is determined according to the importance of each neighboring node to the edge application node.

In this example, as an alternative implementation, the feature corresponding to each node is a text feature of the sequence representation. Therefore, before calculation, each word can also be converted into a vector by using word embedding (word embedding), the dimension of input data is z, the output is set to be a space vector which needs to convert the word into an N (such as set to 64) dimension, the length of the input sequence is F, and therefore the shape of the output data of the layer is (None, F, N). Thus, the vector mapping of the input words is realized, and the index of each word is converted into an N-dimensional fixed shape vector. Alternatively, in determining the attention weight between the edge application node and the neighboring node, the number of convolution kernels of the graph attention layer may be set to 256, and the activation function is set to "relu". It can be 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 scene node may be a converted vector or a text feature before conversion. When the feature is a text feature, before the following operation, the mapping of the text feature is performed, and the text feature is converted into a vector, otherwise, the attention weight is determined directly based on the following mode.

Further, after obtaining the neighboring node feature corresponding to each neighboring node and the second feature, the attention weight between the edge application node and each neighboring node may be determined. For example, after the second feature corresponding to the edge application node and the first feature corresponding to a specific industry client node are obtained, an attention coefficient (attention coefficient) between the edge application node and the industry client node may be calculated first, and then an attention weight (aggregate) between the edge application node and the industry client node is obtained through weighted summation. Then, based on the attention weight, fusing a first feature corresponding to the industry customer node and a second feature corresponding to the edge application node according to the attention weight to obtain a first fused feature; and fusing the third feature and the second feature corresponding to the application scene node according to the attention weight to obtain a second fused feature.

Illustratively, the second characteristic currently determined is the characteristic

The first fusion feature or the second fusion feature is

Further, each node may be applied with a F × F' shared weight matrix W, and then an attention coefficient may be calculated, which may represent the importance of each industry customer node j relative to the edge application node i:

further, to make the attention coefficients easier to calculate and compare, softmax was introduced to regularize all industry client nodes j adjacent to the edge application node i:

wherein alpha is _ij Representing the attention weight between the node pair (i, j), ni represents the set of neighbors of node i, i.e., the set of industry customer nodes.

Further, the first fused feature may be determined as:

it should be noted that, according to the above manner, the second fusion characteristics can be determined as well.

Further, after the first fusion feature and the second fusion feature are determined, the first fusion feature and the second fusion feature may be used as features of target neighboring nodes of the edge application node, and then, based on the above manner, the first fusion feature, the second fusion feature and the second feature are fused to obtain a target feature corresponding to the edge application node. And determining the screening result of the edge application corresponding to the edge application node according to the target characteristics.

Optionally, as an embodiment, referring to fig. 4, the step S40 includes:

and S41, inputting the target features into a pre-trained filter, and determining a filtering result of the edge application through the filter.

In this embodiment, the filter may be trained first based on pre-prepared training data. The training data can collect historical edge application related data sets from the edge computing management module, an edge application heterogeneous graph network which takes an application edge as a center and consists of three heterogeneous nodes of an industry client, an edge application scene and an edge application scene is constructed, the data sets are text features of the three heterogeneous nodes, the three heterogeneous data are subjected to text serialization, and meanwhile, the screening result of each third-party edge application is manually marked.

And respectively coding an industry customer service requirement set, an edge application evaluation set and an edge application scene description set into sequence representations. Defining the length of the coding sequence of each node characteristic as F, taking the longest length F in the data collection as the length of the coding sequence, filling the length of each piece of data as F, and taking the size of a dictionary as z.

Manual marking: and manually marking the screening result of each third-party edge application in each heterogeneous topological graph, wherein the shape of the screening result is N x M (representing the existing M third-party edge applications). Finally, the total data set is divided into training data and test data. For example, 80% of the entire data set may be taken as training data, and the remaining 20% as test data. Training is performed by using a training set, and a verification model is evaluated by using a test set.

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

An error between the predicted edge application filter result and the correct third party edge application filter result is then calculated, with the training objective being to minimize the error. The objective function selects a binary cross entropy (binary _ cross) binary logarithmic loss function. Optionally, the number of training rounds may be set to 1000 (epochs = 1000), and the gradient descent optimization algorithm selects an adam optimizer for improving the learning speed of the conventional gradient descent (optizer = 'adam'). The neural network can find the optimal weight value which enables the target function to be minimum through gradient descent, and the neural network can learn the weight value automatically through training. Training is performed with a training set so that the smaller the objective function, the better, and the validation model is evaluated with a test set after each round of training. And deriving the weight of the model after the model converges.

After the training of the filter is completed, the target feature corresponding to the edge application to be filtered may be determined based on the above steps S10 to S40, and then the filter determines the filtering result of the edge application based on the target feature.

Illustratively, in a specific embodiment, referring to fig. 5, the screening process of the edge application includes:

branch 1: the weight of the importance of the first type neighbor node "industry client" of the edge application node i is learned. Inputting a feature h of a third-party edge application node i _bi And characteristics h of its business customer nodes _pi . Respectively converted into vectors through the word embedding layer and then input into the neighbor node level graph attention layer to obtain a first fusion characteristic h fused with the characteristic information of the neighbor node' industry client _bi1 。

And branch 2: and learning the weight of the second type of neighbor nodes 'application scenes' of the edge application node i on the importance of the second type of neighbor nodes. Inputting the feature h of the edge application node i _bi And applying scene node characteristics h _ui . Respectively converted into vectors through the word embedding layer and then input into the neighbor node level graph attention layer to obtain a second fusion characteristic h fused with the characteristic information of the 'application scene' of the neighbor node _bi2 。

Wherein the word embedding layer (embedding): each word is converted into a vector by word embedding (word embedding), the dimension of input data is z, the output is set to be a space vector which needs to convert the word into 64 dimensions, the length of the input sequence is F, and therefore the shape of the output data of the layer is (None, F, 64). The number of convolution kernels of the neighbor node level graph attention layer is 256, and the activation function is set to be relu.

Further, divide the above two into two partsFirst fused feature h fused with industry client feature and output _bi1 And a second fusion feature h after fusing the application scene feature _bi2 Inputting the data into a central node level graph attention layer for central node level attention aggregation, and outputting a new feature of a first edge application node which is fused with two types of neighbor node features simultaneously, namely a target feature h _bi '. Finally, target feature h _bi ' input to a filter consisting of fully connected layers. To determine the screening result of the edge application by the screener. Wherein, the number of convolution kernels of the central node level graph attention layer is 128, and the activation function is set to be 'relu'.

In the technical scheme disclosed in this embodiment, attention weights between the edge application nodes and each neighboring node are determined, wherein the neighboring nodes include an industry customer node and an application scene node, and then a first feature corresponding to the industry customer node and a second feature corresponding to the edge application node are fused according to the attention weights to obtain a first fused feature; and fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature, determining a target feature corresponding to the edge application node according to the first fused feature and the second fused feature, and finally determining a screening result of the edge application corresponding to the edge application node according to the target feature. Due to the fact that the edge application can be comprehensively screened on the basis of multiple factors of industry customers and edge application scenes, the effect of improving the accuracy of the edge application screening result is achieved.

In addition, an embodiment of the present invention further provides a terminal device, where the terminal device includes: the edge application screening program comprises a memory, a processor and an edge application screening program which is stored on the memory and can run on the processor, wherein the edge application screening program realizes the steps of the edge application screening method according to the various embodiments when being executed by the processor.

In addition, referring to fig. 6, an embodiment of the present invention further provides a terminal device 100, where the terminal device 100 includes:

a determining module 101, configured to determine attention weights between the edge application node and each neighboring node, where the neighboring nodes include an industry customer node and an application scenario node;

a first fusion module 102, configured to fuse, according to the attention weight, a first feature corresponding to the industry client node and a second feature corresponding to the edge application node to obtain a first fusion feature; fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature;

a second fusion module 103, configured to determine, according to the first fusion feature and the second fusion feature, a target feature corresponding to the edge application node;

and the screening module 104 is configured to determine, according to the target feature, a screening result of the edge application corresponding to the edge application node.

Furthermore, an embodiment of the present invention further provides a computer-readable storage medium, where a filter program of an edge application is stored on the computer-readable storage medium, and when the filter program of the edge application is executed by a processor, the steps of the filter method of the edge application described in the above embodiments are implemented.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or system that comprises the element.

The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.

Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as described above and includes several instructions for causing a terminal device (data processing device or server) to execute the method according to the embodiments of the present invention.

The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (10)

1. A screening method for edge application, which is characterized by comprising the following steps:

determining attention weights between the edge application nodes and each adjacent node, wherein the adjacent nodes comprise industry client nodes and application scene nodes;

fusing a first feature corresponding to the industry client node and a second feature corresponding to the edge application node according to the attention weight to obtain a first fused feature; fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature;

determining a target feature corresponding to the edge application node according to the first fusion feature and the second fusion feature;

and determining a screening result of the edge application corresponding to the edge application node according to the target characteristics.

2. The method for screening edge applications according to claim 1, wherein the step of determining attention weights between the edge application nodes and the respective neighboring nodes comprises:

acquiring adjacent node characteristics and the second characteristics corresponding to each adjacent node;

determining the importance of each neighbor node to the edge application node based on a preset shared weight matrix, the neighbor node characteristics corresponding to each neighbor node and the second characteristics;

and determining attention weights between the edge application node and each adjacent node according to the importance of each adjacent node to the edge application node.

3. The method for screening edge applications according to claim 1, wherein the step of determining the screening result of the edge application corresponding to the edge application node according to the target feature comprises:

inputting the target features into a pre-trained filter, and determining a filtering result of the edge application through the filter.

4. The edge applied screening method of claim 3, wherein a binary cross-entropy binary logarithmic loss function is used as an objective function for training the filter during the training of the filter.

5. The method for screening edge applications according to claim 1, wherein the fusing a first feature corresponding to the business customer node and a second feature corresponding to the edge application node according to the attention weight to obtain a first fused feature; before the step of fusing the third feature and the second feature corresponding to the application scene node according to the attention weight to obtain a second fused feature, the method further includes:

acquiring a first text feature corresponding to the industry client node, a second text feature corresponding to the edge application node and a third text feature corresponding to the application scene node;

determining the first feature according to the first text feature, determining the second feature according to the second text feature, and determining the third feature according to the third text feature, wherein the first feature, the second feature, and the third feature are vectors of preset dimensions.

6. The screening method for the edge application according to claim 5, wherein the first text feature comprises a text feature corresponding to an industry customer requirement; the second text features comprise text features corresponding to the edge application evaluation; the third text feature comprises a text feature corresponding to the edge application scene description.

7. The method for screening edge applications according to claim 5, wherein the step of determining attention weights between the edge application node and each neighboring node is preceded by the steps of:

constructing a heteromorphic graph network, wherein the heteromorphic graph network takes the edge application node as a center, and takes the industry customer node and the application scene node as adjacent nodes of the edge application node;

the step of determining the attention weight between the edge application node and each neighboring node comprises:

determining attention weights between the edge application node and respective neighboring nodes based on the heterogeneous graph network.

8. A terminal device, characterized in that the terminal device comprises: a memory, a processor and a filter of an edge application stored on the memory and executable on the processor, the filter of the edge application when executed by the processor implementing the steps of the filtering method of an edge application according to any one of claims 1 to 7.

9. A terminal device, characterized in that the terminal device comprises:

the determining module is used for determining attention weights between the edge application nodes and all adjacent nodes, wherein the adjacent nodes comprise industry client nodes and application scene nodes;

the first fusion module is used for fusing a first feature corresponding to the industry client node and a second feature corresponding to the edge application node according to the attention weight to obtain a first fusion feature; fusing a third feature corresponding to the application scene node and the second feature according to the attention weight to obtain a second fused feature;

the second fusion module is used for determining a target feature corresponding to the edge application node according to the first fusion feature and the second fusion feature;

and the screening module is used for determining the screening result of the edge application corresponding to the edge application node according to the target characteristics.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a filter program of an edge application, which when executed by a processor implements the steps of the filtering method of an edge application according to any one of claims 1 to 7.

Images