Multi-domain data classification method and system based on optimal transmission

Abstract

The invention relates to the technical field of multi-domain data interaction, and discloses a multi-domain data classification method and system based on optimal transmission, and the method comprises the following specific steps: obtaining data with attributes of different domains; preprocessing the data with the attributes, and constructing graph data; considering intra-domain and inter-domain carrying costs and constraint conditions, setting an optimal transmission objective function, and obtaining an inter-domain sample optimal transmission scheme; combining domains by using an inter-domain sample optimal transmission scheme; inputting the merged domains into a graph neural network to obtain node representation; specifically, after a combined domain is input into a graph neural network, each layer of the graph neural network inputs an adjacent matrix and an original attribute of the combined domain into the graph neural network, and node representation is obtained; and obtaining a classification result based on the node representation. The method solves the problem that the existing classification technology neglects the structure information in the edge, and has the characteristics of being capable of improving the data representation learning efficiency and the classification accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-domain data interaction, and more specifically, to a multi-domain data classification method and system based on optimal transmission. Background Art

[0002] Most of the existing graph transfer learning methods are based on domain adaptation. These methods have achieved certain results in the field of graph transfer learning, but there are still some defects. These methods do not consider the potential relationship between samples between domains and cannot encode effective structural information. Moreover, the lack of structural information makes it impossible to fully utilize rich attribute data, which affects the learning of node representation and then affects the accuracy of data prediction.

[0003] There is a graph transfer learning method based on fine-grained domain adaptation, including: collecting samples in the source domain and the target domain, and annotating part or all of the samples in the source domain and the target domain respectively to obtain labeled samples; constructing source domain graphs and target domain graphs for the labeled samples respectively; dividing the labeled samples in the source domain graph into a training set and a validation set, and treating the labeled samples in the target domain graph as a test set; using the training set of the source domain graph and the target domain samples to train the graph neural network to obtain at least two graph neural networks with different parameters; using the validation set of the source domain graph to select the graph neural network; using the selected graph neural network to predict labels for samples in the target domain; by comparing the true labels and predicted labels of all labeled samples in the target domain, obtaining evaluation results; the principle of this method is still based on the traditional domain adaptation method, by reducing the data distribution difference between the two networks, so as to transfer the label information on the existing network to the new network to improve the classification effect. Graph is a special kind of data. In addition to the attributes of the nodes, there is also structural information contained in the edges, and this information is crucial for the representation learning and classification of the nodes, which is a problem that traditional methods are difficult to solve.

[0004] In summary, the existing technology has the problem of ignoring the structural information in the edge. Therefore, how to invent a multi-domain data classification method that takes the edge structure into consideration is a technical problem that needs to be solved urgently in this technical field. Summary of the invention

[0005] In order to solve the problem that the existing classification technology ignores the structural information in the edge, the present invention provides a multi-domain data classification method and system based on optimal transmission, which has the characteristics of being able to improve the data representation learning efficiency and classification accuracy.

[0006] In order to achieve the above-mentioned purpose of the present invention, the technical scheme adopted is as follows:

[0007] A multi-domain data classification method based on optimal transmission includes the following specific steps:

[0008] S1: Get data with attributes in different domains;

[0009] S2: Preprocess the data with attributes and construct graph data;

[0010] S3: Considering the cost and constraints of intra-domain and inter-domain transport, the optimal transport objective function is set to obtain the optimal transport solution for inter-domain samples;

[0011] S4: Merge domains using the optimal transmission scheme of inter-domain samples;

[0012] S5: Input the merged domain into the graph neural network to obtain node representation;

[0013] Specifically, after the merged domain is input into the graph neural network, each layer of the graph neural network inputs the adjacency matrix and original attributes of the merged domain into the graph neural network to obtain the node representation;

[0014] S6: Based on the node representation, obtain the classification result.

[0015] Preferably, in S1, the attributed data is divided into a source domain and a target domain according to the presence or absence of tag information.

[0016] Furthermore, the specific method of preprocessing the data with attributes in S2 and constructing graph data is as follows:

[0017] Treat each sample in the data with attributes as a node;

[0018] If there is a natural pointing relationship between samples, it is used as a basis to connect the two points as an edge; otherwise, the distance between the sample attributes is calculated and used as the weight of the edge between the two points, and finally the edge is stored in the adjacency matrix.

[0019] Furthermore, in step S3, the cost and constraints of intra-domain and inter-domain transport are considered, an optimal transport objective function is set, and an optimal transport solution for inter-domain samples is obtained. The specific steps are as follows:

[0020] S31: Setting the optimal transmission objective function:

[0021] Take the cost and constraints of moving data within and between domains as input;

[0022] The specific transportation costs are:

[0023]

[0024] Among them, C U represents the transport cost within the domain U, C(·,·) represents the distance metric between samples, which is defined as the shortest path between two nodes on the graph. represents the i-th sample in domain U; ​​M represents the transfer cost between domains, and d(·,·) represents the distance metric between attributes, which is defined as the Euclidean distance between two attributes. represents the attribute of the i-th sample in the source domain S, Represents the attribute of the jth sample in the target domain T;

[0025] The specific constraints are:

[0026] μS=π1,μT=π T 1

[0027] Among them, μU represents the contribution of samples in domain U, π represents the optimal transmission scheme, (·) T Represents the transpose of the matrix, 1 is a column vector whose element values ​​are all 1;

[0028] The specific objective function is:

[0029]

[0030] Where π represents the transmission scheme, π* represents the optimal value of the transmission scheme, <·,·> represents the inner product between inputs, α represents the adjustment parameter, and ||·|| represents the L2 norm;

[0031] S32: Optimize the objective function to obtain the optimal transmission solution:

[0032] At step i, calculate the gradient of the objective function to step i-1:

[0033]

[0034] Linear programming problem solving:

[0035]

[0036] Line search obtains weights and updates the transmission plan:

[0037]

[0038] Among them, π (i) is the transmission scheme of step i, τ (i) is the weight of the i-th step.

[0039] Furthermore, in step S4, the domains are merged using the optimal transmission scheme of inter-domain samples, and the specific method is as follows:

[0040] S41: Use the optimal transmission scheme to transfer attributes from the source domain to the target domain as the optimal representation of the source domain data in the target domain, and coordinate with the target domain sample attributes:

[0041] The specific attributes of the transport source domain are:

[0042]

[0043] in, is the attribute of the source domain S sample after transfer, diag(·) -1 Indicates retaining the diagonal elements of the matrix and inverting it, π is the optimal transmission solution, 1 is a column vector with all element values ​​1, X T is the attribute of the sample in the target domain T;

[0044] The merged source domain and target domain attributes are as follows:

[0045]

[0046] in, is the attribute of all samples after merging;

[0047] S42: Using the optimal transmission scheme, the structure is transferred from the source domain to the target domain as the optimal representation of the source domain data in the target domain, and coordinated with the structure between the target domain samples:

[0048] The specific structure of the transport source domain is as follows:

[0049]

[0050] in, is the adjacency matrix of the source domain S sample after transfer, μ S represents the contribution of samples in the source domain S, π is the optimal transmission solution, A T is the adjacency matrix of the target domain T samples, (·) T Represents the transpose of a matrix;

[0051] The structure of merging the source domain and the target domain is as follows:

[0052]

[0053] in, is the adjacency matrix between all samples after merging.

[0054] Furthermore, in step S5, after the merged domain is input into the graph neural network, each layer of the graph neural network inputs the adjacency matrix and original attributes of the merged domain into the graph neural network to obtain a node representation, specifically:

[0055] S51: Input the merged domain into the graph neural network. Each layer of the graph neural network converts each node v From its neighbor nodes Receiving messages:

[0056]

[0057] Among them, AGGREGATE() is an aggregate function. is the neighbor message of node v in the k+1th iteration, is the representation of node i at the kth iteration;

[0058] S52: Each node v updates its features using the received message:

[0059]

[0060] Among them, UPDATE() is the update function;

[0061] After several layers, the final node representation is obtained.

[0062] Furthermore, in S6, based on the node representation, a classification result is obtained, specifically:

[0063] S61: Based on the node representation, a classification algorithm is used to train a classifier, and the trained classifier is used to classify the nodes and edges to be classified to obtain a classification result;

[0064] S62: Use the cross-validation method to verify the classification result of the classifier.

[0065] Furthermore, the aggregate function AGGREGATE() is specifically any one or more functions of sum, average or maximum value.

[0066] Furthermore, the update function UPDATE() is specifically any one or more functions of summation, average or maximum value.

[0067] A multi-domain data classification system based on optimal transmission, comprising the following modules: data acquisition module, preprocessing module, relationship discovery module, collaboration module, graph neural network module, and classification module;

[0068] The data acquisition module is used to obtain data with attributes in different domains;

[0069] The preprocessing module is used to preprocess the data with attributes and construct graph data;

[0070] The relationship discovery module is used to consider the cost and constraints of intra-domain and inter-domain transport, set the optimal transmission objective function, and obtain the optimal transmission solution for inter-domain samples;

[0071] The collaborative module is used to merge domains using an optimal transmission scheme of inter-domain samples;

[0072] The graph neural network module is used to input the merged domain into the graph neural network to obtain node representation; specifically, after the merged domain is input into the graph neural network, each layer of the graph neural network inputs the adjacency matrix and original attributes of the merged domain into the graph neural network to obtain node representation;

[0073] The classification module is used to obtain classification results based on node representation.

[0074] The beneficial effects of the present invention are as follows:

[0075] The present invention processes data to construct graph data; optimal transmission can describe the relationship between attributes of samples and can also mine potential geometric connections; combined with prior knowledge, the relationship between samples is mined through optimal transmission, and the transmission plan is used to merge domains, so that all data can be processed as a whole, thereby improving the efficiency of data representation learning; the graph neural network is used to encode attribute and structural information to obtain a good node representation; based on the node representation, the attribute data to be classified is classified, and the classification results obtained are more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a flow chart of a multi-domain data classification method based on optimal transmission of the present invention.

[0077] Figure 2 It is a system schematic diagram of a multi-domain data classification system based on optimal transmission of the present invention. DETAILED DESCRIPTION

[0078] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0079] Example 1

[0080] like Figure 1 As shown, a multi-domain data classification method based on optimal transmission includes the following specific steps:

[0081] S1: Get data with attributes in different domains;

[0082] S2: Preprocess the data with attributes and construct graph data;

[0083] S3: Considering the cost and constraints of intra-domain and inter-domain transport, the optimal transport objective function is set to obtain the optimal transport solution for inter-domain samples;

[0084] S4: Merge domains using the optimal transmission scheme of inter-domain samples;

[0085] S5: Input the merged domain into the graph neural network to obtain node representation;

[0086] Specifically, after the merged domain is input into the graph neural network, each layer of the graph neural network inputs the adjacency matrix and original attributes of the merged domain into the graph neural network to obtain the node representation;

[0087] S6: Based on the node representation, obtain the classification result.

[0088] Example 2

[0089] More specifically, in a specific embodiment, in S1, the attributed data is divided into a source domain and a target domain according to the presence or absence of tag information.

[0090] Furthermore, the specific method of preprocessing the data with attributes in S2 and constructing graph data is as follows:

[0091] Treat each sample in the data with attributes as a node;

[0092] If there is a natural pointing relationship between samples, it is used as a basis to connect the two points as an edge; otherwise, the distance between the sample attributes is calculated and used as the weight of the edge between the two points, and finally the edge is stored in the adjacency matrix.

[0093] In this embodiment, the distance between sample attributes is specifically a cosine distance.

[0094] In a specific embodiment, in step S3, the cost and constraints of intra-domain and inter-domain transport are considered, an optimal transmission objective function is set, and an optimal transmission scheme for inter-domain samples is obtained, and the specific steps are as follows:

[0095] S31: Setting the optimal transmission objective function:

[0096] Take the cost and constraints of moving data within and between domains as input;

[0097] The specific transportation costs are:

[0098]

[0099] Among them, C U represents the transport cost within the domain U, C(·,·) represents the distance metric between samples, which is defined as the shortest path between two nodes on the graph. represents the i-th sample in domain U; ​​M represents the transfer cost between domains, and d(·,·) represents the distance metric between attributes, which is defined as the Euclidean distance between two attributes. represents the attribute of the i-th sample in the source domain S, Represents the attribute of the jth sample in the target domain T;

[0100] The specific constraints are:

[0101] μS=π1,μT=π T 1

[0102] Among them, μU represents the contribution of samples in domain U, π represents the optimal transmission scheme, (·) T Represents the transpose of the matrix, 1 is a column vector whose element values ​​are all 1;

[0103] The specific objective function is:

[0104]

[0105] Where π represents the transmission scheme, π* represents the optimal value of the transmission scheme, <·,·> represents the inner product between inputs, α represents the adjustment parameter, and ||·|| represents the L2 norm;

[0106] In this embodiment, this objective function is adopted to take into account both the attribute differences between samples and the structural relationships contained between samples, so as to learn more accurate relationships between samples. In addition, by adjusting the weight coefficient, the inclination of attributes and structures is balanced to flexibly adapt to different needs of different data.

[0107] S32: Optimize the objective function to obtain the optimal transmission solution:

[0108] At step i, calculate the gradient of the objective function to step i-1:

[0109]

[0110] Linear programming problem solving:

[0111]

[0112] Line search obtains weights and updates the transmission plan:

[0113]

[0114] Among them, π (i) is the transmission scheme of step i, τ (i) is the weight of the i-th step.

[0115] In this embodiment, by using this optimization objective function, only a small number of non-zero elements are added in each iteration, which can quickly converge and obtain a sparse solution, so that the sample will not find too many samples in another domain, avoiding the problem of over-smoothing in subsequent learning representation, that is, excessive similarity between sample representations;

[0116] In a specific embodiment, in step S4, domains are merged using the optimal transmission scheme for inter-domain samples, and the specific method is:

[0117] S41: Use the optimal transmission scheme to transfer attributes from the source domain to the target domain as the optimal representation of the source domain data in the target domain, and coordinate with the target domain sample attributes:

[0118] The specific attributes of the transport source domain are:

[0119]

[0120] in, is the attribute of the source domain S sample after transfer, diag(·) -1 Indicates retaining the diagonal elements of the matrix and inverting it, π is the optimal transmission solution, 1 is a column vector with all element values ​​1, X T is the attribute of the sample in the target domain T;

[0121] The merged source domain and target domain attributes are as follows:

[0122]

[0123] in, is the attribute of all samples after merging;

[0124] S42: Using the optimal transmission scheme, the structure is transferred from the source domain to the target domain as the optimal representation of the source domain data in the target domain, and coordinated with the structure between the target domain samples:

[0125] The specific structure of the transport source domain is as follows:

[0126]

[0127] in, is the adjacency matrix of the source domain S sample after transfer, μ S represents the contribution of samples in the source domain S, π is the optimal transmission solution, A T is the adjacency matrix of the target domain T samples, (·) T Represents the transpose of a matrix;

[0128] The structure of merging the source domain and the target domain is as follows:

[0129]

[0130] in, is the adjacency matrix between all samples after merging.

[0131] In this embodiment, similarly, the structure between samples in the source domain and the target domain is unified by using the optimal transmission scheme to transport the structure; the purpose of merging domains is to enable samples to utilize the rich information in other domains and fully explore the relationship between samples; and the premise of merging is that the difference between the domain data is as similar as possible to that before the merger. Through the optimal transmission, the data on a certain domain can be "embedded" into the data on another domain to become a whole.

[0132] In a specific embodiment, in step S5, after the merged domain is input into the graph neural network, each layer of the graph neural network inputs the adjacency matrix and original attributes of the merged domain into the graph neural network to obtain a node representation, specifically:

[0133] S51: Input the merged domain into the graph neural network. Each layer of the graph neural network converts each node v From its neighbor nodes Receiving messages:

[0134]

[0135] Among them, AGGREGATE() is an aggregate function. is the neighbor message of node v in the k+1th iteration, is the representation of node i at the kth iteration;

[0136] S52: Each node v updates its features using the received message:

[0137]

[0138] Among them, UPDATE() is the update function;

[0139] After several layers, the final node representation is obtained.

[0140] In a specific embodiment, in S6, the classification result is obtained based on the node representation, specifically:

[0141] S61: Based on the node representation, a classification algorithm is used to train a classifier, and the trained classifier is used to classify the nodes and edges to be classified to obtain a classification result;

[0142] S62: The classification result of the classifier is verified by the cross-validation method, and the role of the cross-validation method is to prevent overfitting.

[0143] In a specific embodiment, the aggregation function AGGREGATE() is specifically any one or more functions of sum, average or maximum value.

[0144] In a specific embodiment, the update function UPDATE() is specifically any one or more functions of summation, average or maximum value.

[0145] Example 3

[0146] like Figure 2 As shown, a multi-domain data classification system based on optimal transmission includes the following modules: data acquisition module, preprocessing module, relationship discovery module, collaboration module, graph neural network module, and classification module;

[0147] The data acquisition module is used to obtain data with attributes in different domains;

[0148] The preprocessing module is used to preprocess the data with attributes and construct graph data;

[0149] The relationship discovery module is used to consider the cost and constraints of intra-domain and inter-domain transport, set the optimal transmission objective function, and obtain the optimal transmission solution for inter-domain samples;

[0150] The collaborative module is used to merge domains using an optimal transmission scheme of inter-domain samples;

[0151] The graph neural network module is used to input the merged domain into the graph neural network to obtain node representation; specifically, after the merged domain is input into the graph neural network, each layer of the graph neural network inputs the adjacency matrix and original attributes of the merged domain into the graph neural network to obtain node representation;

[0152] The classification module is used to obtain classification results based on node representation.

[0153] The relationship discovery module includes a first step unit and a second step unit;

[0154] The first step unit is used to set the optimal transmission objective function and define its input and mathematical expression:

[0155] The cost and constraints of data transfer within and between domains are used as input; the transfer cost is:

[0156]

[0157] Among them, C U represents the transport cost within the domain U, C(·,·) represents the distance metric between samples, which is defined as the shortest path between two nodes on the graph. represents the i-th sample in domain U; ​​M represents the transfer cost between domains, and d(·,·) represents the distance metric between attributes, which is defined as the Euclidean distance between two attributes. represents the attribute of the i-th sample in the domain S, represents the attribute of the jth sample in the domain T;

[0158] The constraints are:

[0159] μS=T1,μT=T T 1

[0160] Where μU represents the contribution of samples in domain U, T represents the optimal transmission scheme, (·) T Represents the transpose of the matrix, 1 is a column vector whose element values ​​are all 1;

[0161] The objective function is:

[0162]

[0163] Where π represents the transmission scheme, π* represents the optimal value of the transmission scheme, <·,·> represents the inner product between inputs, α represents the adjustment parameter, and ||·|| represents the L2 norm;

[0164] Used to optimize the objective function and obtain the optimal transmission solution:

[0165] At step i, calculate the gradient of the objective function to step i-1:

[0166]

[0167] Linear programming problem solving:

[0168]

[0169] Line search obtains weights and updates the transmission plan:

[0170]

[0171] Among them, τ( i ) is the weight of the i-th step, π ( i ) is the transmission scheme of step i.

[0172] The collaboration module includes an attribute unit and a structure unit;

[0173] Attribute unit: Use the optimal transmission scheme to transfer attributes from the source domain to the target domain as the optimal representation of the source domain data in the target domain, and coordinate with the target domain sample attributes:

[0174] First, move the source domain attributes:

[0175]

[0176] in, is the attribute of the source domain S sample after transfer, dia.g(·) -1Indicates retaining the diagonal elements of the matrix and inverting it, π is the optimal transmission solution, 1 is a column vector with all element values ​​1, X T is the attribute of the sample in the target domain T;

[0177] Merge source and target domain attributes:

[0178]

[0179] in, is the attribute of all samples after merging;

[0180] Structural unit: Use the optimal transmission scheme to transfer the structure from the source domain to the target domain as the optimal representation of the source domain data in the target domain, and coordinate with the structure between the target domain samples:

[0181] First, move the source domain structure:

[0182]

[0183] in, is the adjacency matrix of the source domain S sample after transfer, μ S represents the contribution of samples in the source domain S, π is the optimal transmission solution, A T is the adjacency matrix of the target domain T samples, (·) T Represents the transpose of a matrix;

[0184] Merge the source and target domain structures:

[0185]

[0186] in, is the adjacency matrix between all samples after merging.

[0187] The classification module includes a classification unit and a verification unit;

[0188] The classification unit uses the classification algorithm to train the classifier based on the node representation, and uses the trained classifier to classify the nodes and edges to be classified to obtain the classification results;

[0189] The verification unit uses the cross-validation method to verify the classification results of the classifier.

[0190] Obviously, the above embodiments of the present invention are only examples for clearly explaining the present invention, and are not intended to limit the implementation methods of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. A multi-domain data classification method based on optimal transmission, characterized in that: The specific steps include: S1: Get data with attributes in different domains; S2: Preprocess the data with attributes and construct graph data; S3: Considering the cost and constraints of intra-domain and inter-domain transport, the optimal transport objective function is set to obtain the optimal transport solution for inter-domain samples; S4: Merge domains using the optimal transmission scheme of inter-domain samples; S5: Input the merged domain into the graph neural network to obtain node representation; Specifically, after the merged domain is input into the graph neural network, each layer of the graph neural network inputs the adjacency matrix and original attributes of the merged domain into the graph neural network to obtain the node representation; S6: Based on the node representation, obtain the classification result.

2. The multi-domain data classification method based on optimal transmission according to claim 1 is characterized in that: In S1, the attributed data is divided into a source domain and a target domain according to the presence or absence of tag information.

3. The multi-domain data classification method based on optimal transmission according to claim 2 is characterized in that: In S2, the data with attributes is preprocessed, and the specific method of constructing graph data is as follows: Treat each sample in the data with attributes as a node; If there is a natural pointing relationship between samples, it is used as a basis to connect the two points as an edge; otherwise, the distance between the sample attributes is calculated and used as the weight of the edge between the two points, and finally the edge is stored in the adjacency matrix.

4. The multi-domain data classification method based on optimal transmission according to claim 3 is characterized in that: In step S3, the cost and constraints of intra-domain and inter-domain transport are considered, an optimal transport objective function is set, and an optimal transport solution for inter-domain samples is obtained. The specific steps are as follows: S31: Setting the optimal transmission objective function: Take the cost and constraints of moving data within and between domains as input; The specific transportation cost is: Among them, C U represents the transport cost within the domain U, C(·,·) represents the distance metric between samples, which is defined as the shortest path between two nodes on the graph. represents the i-th sample in domain U; ​​M represents the transfer cost between domains, and d(·,·) represents the distance metric between attributes, which is defined as the Euclidean distance between two attributes. represents the attribute of the i-th sample in the source domain S, Represents the attribute of the jth sample in the target domain T; The specific constraints are: m S =π1,μ T =π T 1 Among them, μ U represents the contribution of samples in domain U, π represents the optimal transmission scheme, (·) T Represents the transpose of the matrix, 1 is a column vector whose element values ​​are all 1; The specific objective function is: Among them, π represents the transmission scheme, π * represents the optimal value of the transmission scheme, 〈·,·〉 represents the inner product between inputs, α represents the adjustment parameter, and ‖·‖ represents the L2 norm; S32: Optimize the objective function to obtain the optimal transmission solution: At step i, calculate the gradient of the objective function to step i-1: Linear programming problem solving: Line search obtains weights and updates the transmission plan: Among them, π (i) is the transmission scheme of step i, τ (i) is the weight of the i-th step.

5. The multi-domain data classification method based on optimal transmission according to claim 4 is characterized in that: In step S4, the domains are merged using the optimal transmission scheme for inter-domain samples. The specific method is as follows: S41: Use the optimal transmission scheme to transfer attributes from the source domain to the target domain as the optimal representation of the source domain data in the target domain, and coordinate with the target domain sample attributes: The specific attributes of the transport source domain are: in, is the attribute of the source domain S sample after transfer, diag(·) -1 Indicates retaining the diagonal elements of the matrix and inverting it, π is the optimal transmission solution, 1 is a column vector with all element values ​​1, X T is the attribute of the sample in the target domain T; The merged source domain and target domain attributes are as follows: in, is the attribute of all samples after merging; S42: Using the optimal transmission scheme, the structure is transferred from the source domain to the target domain as the optimal representation of the source domain data in the target domain, and coordinated with the structure between the target domain samples: The specific structure of the transport source domain is: in, is the adjacency matrix of the source domain S sample after transfer, μ S represents the contribution of samples in the source domain S, π is the optimal transmission solution, A T is the adjacency matrix of the target domain T samples, (·) T Represents the transpose of a matrix; The structure of merging the source domain and the target domain is as follows: in, is the adjacency matrix between all samples after merging.

6. The multi-domain data classification method based on optimal transmission according to claim 5 is characterized in that: In step S5, after the merged domain is input into the graph neural network, each layer of the graph neural network inputs the adjacency matrix and original attributes of the merged domain into the graph neural network to obtain a node representation, specifically: S51: The merged domain is input into the graph neural network. Each layer of the graph neural network separates each node v from its neighbor nodes Receiving messages: Among them, AGGREGATE() is an aggregate function. is the neighbor message of node v in the k+1th iteration, is the representation of node i at the kth iteration; S52: Each node v updates its features using the received message: Among them, UPDATE() is the update function; After several layers, the final node representation is obtained.

7. The multi-domain data classification method based on optimal transmission according to claim 6 is characterized in that: In S6, based on the node representation, a classification result is obtained, specifically: S61: Based on the node representation, a classification algorithm is used to train a classifier, and the trained classifier is used to classify the nodes and edges to be classified to obtain a classification result; S62: Use the cross-validation method to verify the classification result of the classifier.

8. The multi-domain data classification method based on optimal transmission according to claim 6 is characterized in that: The aggregation function AGGREGATE() is specifically any one or more functions of sum, average or maximum value.

9. The multi-domain data classification method based on optimal transmission according to claim 6, characterized in that: The update function UPDATE() is specifically any one or more functions of summation, average or maximum value.

10. A multi-domain data classification system based on optimal transmission, characterized in that: It includes the following modules: data acquisition module, preprocessing module, relationship discovery module, collaboration module, graph neural network module, and classification module; The data acquisition module is used to obtain data with attributes in different domains; The preprocessing module is used to preprocess the data with attributes and construct graph data; The relationship discovery module is used to consider the cost and constraints of intra-domain and inter-domain transport, set the optimal transmission objective function, and obtain the optimal transmission solution for inter-domain samples; The collaborative module is used to merge domains using an optimal transmission scheme of inter-domain samples; The graph neural network module is used to input the merged domain into the graph neural network to obtain node representation; specifically, after the merged domain is input into the graph neural network, each layer of the graph neural network inputs the adjacency matrix and original attributes of the merged domain into the graph neural network to obtain node representation; The classification module is used to obtain classification results based on node representation.

Images