A graph classification test model data set testing method, terminal and storage medium

Abstract

The present invention relates to the field of data testing technology, and discloses a graph classification test model data set testing method, terminal and storage medium, the method comprising: obtaining a data set to be evaluated, and obtaining multiple graph classification test model classes, each graph classification test model class including more than two graph classification test models, inputting the data set to be evaluated into each graph classification test model respectively, and obtaining the test index corresponding to each graph classification test model; calculating the performance difference corresponding to each graph classification test model class according to the test index corresponding to each graph classification test model; calculating the data set performance index of the data set to be evaluated according to the performance difference corresponding to each graph classification test model class. The present invention calculates the performance index of the data set to be evaluated to provide more references to testers, help testers select appropriate data sets, and is used to clearly distinguish the progress of different models in subsequent tests and help select appropriate models.

Description

Technical Field

[0001] The present invention relates to the field of data testing technology, and in particular to a graph classification test model data set testing method, terminal and storage medium. Background Art

[0002] In the field of graph classification, Graph Neural Networks (GNNs), as an effective machine learning method, has been widely used in many fields, such as recommendation systems, molecular property prediction, and natural language processing. In order to evaluate the performance of graph neural network models in these tasks, specific datasets are often selected as benchmark datasets. Ideal benchmark datasets should be able to clearly distinguish the progress of different models.

[0003] Although existing graph benchmarking frameworks provide a standard model performance evaluation platform for graph classification tasks, in tests, graph neural networks sometimes do not outperform other baseline methods in all graph classification tasks. Some simple baseline methods, such as MLPs (multi-layer perceptrons), can sometimes even achieve comparable or better performance than graph neural networks, which violates the theoretical performance of graph neural networks and means that many datasets used to evaluate graph neural networks in the existing technology are inappropriate and cannot clearly distinguish the progress of different models.

[0004] To this end, a method is needed to evaluate the performance of different data sets to help testers select appropriate data sets, so as to clearly distinguish the progress of different models in subsequent tests and help select appropriate models. Summary of the invention

[0005] The purpose of the present invention is to provide a graph classification test model dataset testing method, aiming to solve the problem in the prior art that the test dataset of the graph classification model lacks evaluation criteria and is prone to selecting inappropriate datasets.

[0006] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0007] The present invention provides a graph classification test model data set testing method, the graph classification test model data set testing method comprising:

[0008] Obtain a data set to be evaluated, and obtain multiple graph classification test model classes, each of which includes more than two graph classification test models, input the data set to be evaluated into each graph classification test model, and obtain test indicators corresponding to each graph classification test model;

[0009] According to the test indicators corresponding to each graph classification test model, calculate the performance difference corresponding to each graph classification test model class;

[0010] The dataset performance index of the dataset to be evaluated is calculated based on the performance differences corresponding to each graph classification test model class.

[0011] Furthermore, the obtaining of the data set to be evaluated specifically includes:

[0012] Get A collection of tags and the number of labels C, and specify each label Attribute distribution , get the attribute distribution set of the label ,in Indicates the tag number;

[0013] For each tag , N samples are sampled to meet the attribute distribution A sample collection ;

[0014] Based on the collected sample set Update attribute distribution ;

[0015] Distribute collections based on label attributes The collection of all samples Calculate and get the classified category set ;

[0016] Classification set for classification Perform normalization:

[0017] ;

[0018] in, is the paradigm function, is the rounding function, is the number of categories for classification;

[0019] Classification by category , a collection of samples Constructing the evaluation dataset ,in, For the samples, For the The labels corresponding to the samples.

[0020] Furthermore, the tag Attribute distribution Is a normal distribution and uniform distribution One of;

[0021] If the label Attribute distribution Is a normal distribution , then the sample set collected according to Update attribute distribution Specifically, according to the collected samples , and the normal distribution First parameter and the second parameter Update attribute distribution :

[0022] ;

[0023] in, represents the assignment operator;

[0024] If the label Attribute distribution Is uniformly distributed , then the sample set collected according to Update attribute distribution Specifically, according to the collected samples , and the normal distribution First parameter and the second parameter Update attribute distribution :

[0025] .

[0026] Furthermore, the graph classification test model class includes a structural feature class and a node attribute class, the structural feature class includes a first structural graph classification test model and a second structural graph classification test model, and the node attribute class includes a first node graph classification test model and a second node graph classification test model;

[0027] The step of inputting the to-be-evaluated data set into each of the graph classification test models to obtain the test indicators corresponding to each of the graph classification test models specifically includes:

[0028] The data set to be evaluated is input into the first structure graph classification test model to obtain the first structure test index , input the data set into the second structure graph classification test model to obtain the second structure test index ;

[0029] Input the data set into the first node graph classification test model to obtain the first node test index , input the data set into the second node graph classification test model to obtain the second node test index .

[0030] Furthermore, the performance differences corresponding to the various graph classification test model classes are calculated according to the test indicators corresponding to the various graph classification test models, specifically including:

[0031] According to the first structural test index and the second structure test index , calculate the first performance difference corresponding to the structural feature class:

[0032] ;

[0033] in, The dataset to be evaluated The first performance difference;

[0034] According to the first node test index and the second node test index , calculate the second performance difference corresponding to the node attribute class:

[0035] ;

[0036] in, The dataset to be evaluated The second performance difference.

[0037] Furthermore, the data set performance index of the data set to be evaluated is calculated according to the performance differences corresponding to each graph classification test model class, specifically including:

[0038] The performance differences corresponding to each graph classification test model class are standardized to obtain the standardized performance differences corresponding to each graph classification test model class:

[0039] ;

[0040] in, The dataset to be evaluated Testing model classes in graph classification The performance of the standard after normalization is poor, The dataset to be evaluated Testing model classes in graph classification The performance difference, It is a graph classification test model class The smallest test index among all the test indicators, is the number of categories for classification;

[0041] According to the standardized performance difference corresponding to each graph classification test model class, a single performance evaluation index corresponding to each graph classification test model class is calculated.

[0042] Furthermore, the single performance evaluation index corresponding to each graph classification test model class is calculated according to the standardized performance difference corresponding to each graph classification test model class, specifically including:

[0043] Calculate the complexity factor corresponding to each graph classification test model class:

[0044] ;

[0045] in, is the complexity factor;

[0046] The single performance evaluation index is obtained by multiplying the standardized performance difference and complexity factor corresponding to each graph classification test model class.

[0047] Furthermore, the calculation of the data set performance index of the data set to be evaluated according to the performance differences corresponding to each graph classification test model class also includes:

[0048] The data set performance index is obtained by summing up the single performance evaluation indexes:

[0049] ;

[0050] in, Represents the graph classification test model class A single performance evaluation indicator, Represents the dataset to be evaluated The performance index of the dataset is Represents the structural feature class, Represents a node attribute class.

[0051] In addition, to achieve the above-mentioned purpose, the present invention also provides a storage medium, which stores a graph classification test model dataset testing program, and when the graph classification test model dataset testing program is executed by a processor, it implements the steps of the graph classification test model dataset testing method as described above.

[0052] In addition, to achieve the above-mentioned purpose, the present invention also provides a readable storage medium, which stores a graph classification test model dataset testing program, and when the graph classification test model dataset testing program is executed by a processor, it implements the steps of the graph classification test model dataset testing method as described above.

[0053] The present invention adopts the above technical solution to achieve the following effects:

[0054] The present invention adopts the test set to be evaluated to test multiple graph classification test models of multiple graph classification test model classes respectively to obtain the test indicators of different graph classification test models under different graph classification test model classes, and then evaluates the difference in test indicators of different graph classification test models under a single graph classification test model class, so as to obtain the ability of the data set to be evaluated to distinguish the classification effects of different models under the graph classification test model class, so as to characterize the effect of the data set to be evaluated under the graph classification test model class, and then, according to the performance difference of the data set to be evaluated under the corresponding graph classification test model class, further calculate the performance indicator of the data set to be evaluated, so as to provide more references to testers and help testers select suitable data sets, so as to clearly distinguish the progress of different models and help select suitable models in subsequent tests. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flowchart of the steps of a method for testing a graph classification test model data set in a preferred embodiment of the present invention;

[0056] Figure 2 It is a structural flow chart of a graph classification test model data set testing method in a preferred embodiment of the present invention;

[0057] Figure 3 is a schematic diagram of the properties of the data set performance indicator in a preferred embodiment of the present invention;

[0058] Figure 4 are evaluation results of different data sets in a preferred embodiment of the present invention;

[0059] Figure 5 It is a schematic diagram of the operating environment of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments. The specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention. All other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.

[0061] Embodiment 1

[0062] See also Figure 1 and Figure 2 , Figure 1 is a flowchart of the steps of a graph classification test model data set testing method in a preferred embodiment of the present invention, Figure 2It is a structural flow chart of a graph classification test model data set testing method in a preferred embodiment of the present invention. Embodiment 1 of the present application is a graph classification test model data set testing method, which includes the steps of:

[0063] S1. Obtain a data set to be evaluated, and obtain multiple graph classification test model classes, each of which includes more than two graph classification test models, and input the data set to be evaluated into each graph classification test model to obtain the test indicators corresponding to each graph classification test model.

[0064] The present invention takes into account that existing graph classification benchmark datasets mostly focus on the single use of structural or attribute information, fail to fully utilize the comprehensive characteristics of graph data, and usually ignore the systematic definition and measurement of the effectiveness of the dataset in the evaluation framework, so that the results of the benchmark test may not truly reflect the advantages and disadvantages of different graph neural network models. More importantly, the existing benchmark tests do not consider the correlation between the intrinsic attributes of the dataset and the classification labels, which may cause some baseline methods to perform well due to accidental attribute label associations, thereby masking the potential advantages of more complex models. For example, if the degree distribution of nodes in a dataset is highly correlated with the class label, even a simple degree-based classification method may show performance comparable to that of a complex GNN model. For this reason, the present invention uses multiple categories of graph classification test models to test the dataset to be evaluated, so as to obtain multiple categories of test indicators, and make judgments on the evaluation ability of the graph classification test model to be evaluated for multiple categories of models.

[0065] Specifically, in this embodiment, the graph classification test model class includes a structural feature class and a node attribute class, wherein the structural feature class refers to the category of the graph classification test model that uses structural features as model input, and the node attribute class refers to the category of the graph classification test model that uses node attribute features as model input.

[0066] In this embodiment, the structural feature class includes two graph classification test models, namely, a first structural graph classification test model and a second structural graph classification test model, and the node attribute class also includes two graph classification test models, namely, a first node graph classification test model and a second node graph classification test model.

[0067] The data set is input into the first structure graph classification test model to obtain the first structure test index ; Input the second structure diagram classification test model to obtain the second structure test index ; Input the first node graph classification test model to obtain the first node test index ; Input the second node graph classification test model to obtain the second node test index .

[0068] In this embodiment, the test index can be one or more of the classification performance indicators such as AUC-ROC (Area Under the Receiver Operating Characteristic Curve) or accuracy. It is set according to actual needs. If it is necessary to measure the ability of the data set to be evaluated to distinguish different models in terms of accuracy, the test index can select the accuracy index. If it is necessary to comprehensively measure the ability of the data set to distinguish different models, the test index can select multiple indicators.

[0069] In this embodiment, the dataset to be evaluated Specifically, a controllable synthetic data set is generated to obtain the data set to be evaluated. That is, a controllable synthetic dataset is generated, which includes:

[0070] Get A collection of tags and the number of labels C, and specify the attribute distribution of each label Among them, the attribute distribution can be a normal distribution or evenly distributed .

[0071] Afterwards, based on the collection of tags , the number of labels C and the attribute distribution of each label Construct a dataset of size N to be evaluated :

[0072] First, for each label ,all:

[0073] Take N samples , these samples obey the attribute distribution of the label ,in Represents the label number, which in this embodiment can come from a normal distribution or evenly distributed , the sample size is N;

[0074] Update label attribute distribution , if a normal distribution is adopted, then:

[0075] ;

[0076] in, Is the assignment symbol.

[0077] If a uniform distribution is used, then:

[0078] ;

[0079] Afterwards, the set of all samples Calculate and get the classified category set ;

[0080] Then, the classification category set Perform normalization:

[0081] ;

[0082] in, is the paradigm function, Right now , It is a rounding function that controls the number of classification categories through standardization. In the range of the number of tags C.

[0083] Finally, construct the final dataset to be evaluated , among which Sample graph With attributes , the corresponding label Group by Category Sure.

[0084] S2. According to the test indicators corresponding to each graph classification test model, calculate the performance difference corresponding to each graph classification test model class.

[0085] Specifically, in this embodiment, the performance differences of the graph classification test models of the structural feature class and the node attribute class are calculated respectively.

[0086] The dataset to be evaluated The first performance difference of the graph classification test model of the structural feature class is:

[0087] ;

[0088] in, The dataset to be evaluated The first performance difference of . It represents the ability difference of measuring the classification performance of different models in the graph classification test model with structural features as input of the evaluation dataset.

[0089] Dataset to be evaluated The second performance difference of the graph classification test model of the node attribute class is:

[0090] ;

[0091] in, The dataset to be evaluated The second performance difference represents the ability difference of the evaluation data set to measure the classification performance of different models in the graph classification test model with node attribute features as input.

[0092] In this embodiment, by generating a controllable synthetic data set, the user is allowed to independently generate and adjust the structure and attribute characteristics of the data set. In the later process, based on the test and evaluation results of the data set, the controllable synthetic data set is finely adjusted in a targeted manner, so as to generate a graph data set with specific attributes and structural characteristics according to preset parameters, so as to be used in a targeted manner to evaluate the performance of the model.

[0093] S3. Calculate the dataset performance index of the dataset to be evaluated based on the performance differences corresponding to each graph classification test model class.

[0094] Specifically, in this embodiment, the performance difference is first standardized:

[0095] ;

[0096] in, The dataset to be evaluated The performance of the standard after normalization is poor, , The dataset to be evaluated Poor performance, include and , The dataset to be evaluated The first performance difference, The dataset to be evaluated The second performance difference, is a collection of categories, is the number of categories for classification, poor performance Divide by Standardization, where It is the worse performance indicator of the two methods. In this embodiment, when hour, ,when hour, This ratio takes into account the number of categories Normalizing the performance impact allows for fair comparisons between different datasets.

[0097] Afterwards, the complexity factor is used to evaluate the normalized performance difference:

[0098] ;

[0099] in, is the complexity factor. In the above formula, the denominator is the accuracy of random guessing. The higher the value of , the more complex the dataset is, that is, the lower the performance of the baseline method. This definition not only provides a new perspective to evaluate the quality and applicability of a dataset, but also promotes the development of more scientific and comprehensive benchmarking methods.

[0100] Afterwards, the standardized performance difference is multiplied by the complexity factor to obtain a single performance evaluation index corresponding to the graph classification test model class. Finally, the single performance evaluation indexes of all graph classification test model classes are accumulated to obtain the data set performance index:

[0101] ;

[0102] in, Is the dataset to be evaluated The performance indicators of the dataset.

[0103] Dataset performance indicators , which is used to quantify the ability of graph classification datasets to distinguish the performance of different models. This indicator is based on the performance differences of the models on the dataset, combined with the characteristics of the dataset itself, such as category distribution, graph structure, and complexity of attribute features. The value range of is [0, 1]. The larger the value, the higher the validity of the data set. This paper verifies the validity and scientificity of this indicator by conducting a detailed analysis of the complexity factor. Figure 3 As shown, taking the structural feature graph classification test model class as an example, this complexity factor has good effectiveness and has the following four properties:

[0104] Property 1: When the accuracy of the worst method improves, Linearly decrease. Figure 3 In (a), 9 curves of different colors are shown, each curve represents the corresponding dataset with different number of task labels (from 2 to 10). Value, fixed =1, and let from changes to 1. Obviously, as The increase of all curves The values ​​decrease linearly.

[0105] Property 2: When the performance gap is reduced (improved Will cause fixation The performance gap is reduced), data set performance indicators It will also decrease, such as Figure 3 As shown in (b), set from Change to 1, =1, poor performance From 1- Changes to 0.

[0106] Property 3: For a given dataset, if the relative performance difference between the two methods remains unchanged, the dataset performance index The value should be based on the absolute performance values ​​of the two methods ( , ) changes. Specifically, higher absolute performance values ​​should be associated with smaller data set performance indicators For example, consider a performance difference of 0.1 between two methods. In a binary classification scenario, when =0.5, =0.6 Should be greater than =0.9, =1.0 This is because lower absolute performance indicates that the dataset is non-trivial, thus yielding a larger .like Figure 3 As shown in (c), the performance difference is set to 0.2. from Change to 0.8, Follow decreases with the increase of .

[0107] Property 4: If two data sets and have the same performance difference, and Label set Size is less than Label set ,So Dataset performance indicators Dataset performance indicators . Figure 3 This property is shown as smaller The curve is located at a larger above the curve.

[0108] Please refer to Figure 4 , Figure 4 The evaluation results on existing real data sets are given. Figure 4 (a) in the figure is the evaluation result of different data sets using accuracy as the evaluation indicator. Figure 4 (b) in the figure is the evaluation result of different data sets using the receiver operating characteristic curve and the area under the curve as evaluation indicators.

[0109] Embodiment 2

[0110] See also Figure 5Based on the above method, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 5 Only some components of the terminal are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0111] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Furthermore, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a graph classification test model data set test program 40 is stored on the memory 20, and the graph classification test model data set test program 40 can be executed by the processor 10.

[0112] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run the program code or process data stored in the memory 20, such as executing relevant programs of the graph classification test model data set testing method.

[0113] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 30 is used to display information on the terminal and to display a visual user interface. The components of the terminal communicate with each other via a system bus.

[0114] In one embodiment, when the processor 10 executes the graph classification test model dataset testing program 40 in the memory 20, the steps of a graph classification test model dataset testing method as described above are implemented.

[0115] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), etc. equipped on the terminal. Furthermore, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a graph classification test model data set test program is stored on the memory 20, and the graph classification test model data set test program can be executed by the processor 10, thereby realizing the graph classification test model data set test method in the present application.

[0116] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run the program code or process data stored in the memory 20, such as executing the graph classification test model data set testing method.

[0117] Embodiment 3

[0118] This embodiment provides a storage medium, which stores a graph classification test model dataset testing program. When the graph classification test model dataset testing program is executed by a processor, the steps of the graph classification test model dataset testing method as described above are implemented.

[0119] In summary, the present invention adopts the test set to be evaluated to test multiple graph classification test models of multiple graph classification test model classes respectively to obtain the test indicators of different graph classification test models under different graph classification test model classes, and then evaluates the difference in test indicators of different graph classification test models under a single graph classification test model class, so as to obtain the ability of the data set to be evaluated to distinguish the classification effects of different models under the graph classification test model class, so as to characterize the effect of the data set to be evaluated under the graph classification test model class, and then, according to the performance difference of the data set to be evaluated under the corresponding graph classification test model class, further calculate the performance indicator of the data set to be evaluated, so as to provide more references to testers and help testers select suitable data sets, so as to clearly distinguish the progress of different models and help select suitable models in subsequent tests.

[0120] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or terminal including the element.

[0121] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable storage medium, and the program can include the processes of the above-mentioned method embodiments when executed. The storage medium can be a memory, a disk, an optical disk, etc.

[0122] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A graph classification test model dataset testing method, characterized in that: The graph classification test model data set testing method comprises: Obtain a data set to be evaluated, and obtain multiple graph classification test model classes, each of which includes more than two graph classification test models, input the data set to be evaluated into each graph classification test model, and obtain test indicators corresponding to each graph classification test model; According to the test indicators corresponding to each graph classification test model, calculate the performance difference corresponding to each graph classification test model class; Calculate the dataset performance index of the dataset to be evaluated based on the performance differences corresponding to each graph classification test model class; The graph classification test model class includes a structural feature class and a node attribute class, the structural feature class includes a first structural graph classification test model and a second structural graph classification test model, and the node attribute class includes a first node graph classification test model and a second node graph classification test model; The step of inputting the to-be-evaluated data set into each of the graph classification test models to obtain the test indicators corresponding to each of the graph classification test models specifically includes: The data set to be evaluated is input into the first structure graph classification test model to obtain the first structure test index , input the data set into the second structure graph classification test model to obtain the second structure test index ; Input the data set into the first node graph classification test model to obtain the first node test index , input the data set into the second node graph classification test model to obtain the second node test index .

2. A graph classification test model dataset testing method according to claim 1, characterized in that: The obtaining of the data set to be evaluated specifically includes: Get A collection of tags and the number of labels C, and specify each label Attribute distribution , get the attribute distribution set of the label ,in Indicates the tag number; For each tag , N samples are sampled to meet the attribute distribution A sample collection ; Based on the collected sample set Update attribute distribution ; Distribute collections based on label attributes The collection of all samples Calculate and get the classified category set ; Classification set for classification Perform normalization: ； in, is the paradigm function, is the rounding function, is the number of categories for classification; Classification by category , a collection of samples Constructing the evaluation dataset ,in, For the samples, For the The labels corresponding to the samples.

3. A graph classification test model data set testing method according to claim 2, characterized in that: Label Attribute distribution Is a normal distribution and uniform distribution One of; If the label Attribute distribution Is a normal distribution , then the sample set collected according to Update attribute distribution Specifically, according to the collected samples , and the normal distribution First parameter and the second parameter Update attribute distribution : ； in, represents the assignment operator; If the label Attribute distribution Is uniformly distributed , then the sample set collected according to Update attribute distribution Specifically, according to the collected samples , and the normal distribution First parameter and the second parameter Update attribute distribution : 。 4. The method for testing a graph classification test model dataset according to claim 1, characterized in that: The method of calculating the performance difference corresponding to each graph classification test model class according to the test indicators corresponding to each graph classification test model specifically includes: According to the first structural test index and the second structure test index , calculate the first performance difference corresponding to the structural feature class: ； in, The dataset to be evaluated The first performance difference; According to the first node test index and the second node test index , calculate the second performance difference corresponding to the node attribute class: ； in, The dataset to be evaluated The second performance difference.

5. A graph classification test model data set testing method according to claim 4, characterized in that: The calculation of the data set performance index of the data set to be evaluated according to the performance differences corresponding to each graph classification test model class specifically includes: The performance differences corresponding to each graph classification test model class are standardized to obtain the standardized performance differences corresponding to each graph classification test model class: ； in, The dataset to be evaluated Testing model classes in graph classification The performance of the standard after normalization is poor, The dataset to be evaluated Testing model classes in graph classification The performance difference, It is a graph classification test model class The smallest test index among all the test indicators, is the number of categories for classification; According to the standardized performance difference corresponding to each graph classification test model class, a single performance evaluation index corresponding to each graph classification test model class is calculated.

6. A graph classification test model data set testing method according to claim 5, characterized in that: The method of calculating a single performance evaluation index corresponding to each graph classification test model class according to the standardized performance difference corresponding to each graph classification test model class specifically includes: Calculate the complexity factor corresponding to each graph classification test model class: ； in, is the complexity factor; The single performance evaluation index is obtained by multiplying the standardized performance difference and complexity factor corresponding to each graph classification test model class.

7. A graph classification test model data set testing method according to claim 6, characterized in that: The step of calculating the data set performance index of the data set to be evaluated according to the performance differences corresponding to each graph classification test model class also includes: The data set performance index is obtained by summing up the single performance evaluation indexes: ； in, Represents the graph classification test model class A single performance evaluation indicator, Represents the dataset to be evaluated The performance index of the dataset is Represents the structural feature class, Represents a node attribute class.

8. A terminal, characterized in that: The terminal includes: a memory, a processor, and a graph classification test model dataset test program stored in the memory and executable on the processor. When the graph classification test model dataset test program is executed by the processor, the terminal is controlled to implement the steps of the graph classification test model dataset test method as described in any one of claims 1-7.

9. A storage medium, characterized in that: The storage medium stores a graph classification test model dataset testing program, and when the graph classification test model dataset testing program is executed by a processor, the steps of the graph classification test model dataset testing method as described in any one of claims 1-7 are implemented.

Images