Financial data anomaly detection method based on multi-channel graph contrast learning

Abstract

The invention relates to a financial data anomaly detection method based on multi-channel graph contrast learning, and belongs to the field of financial data anomaly detection. According to the method, a financial information relation graph with abnormity is input, and abnormal information and behaviors of each user in the graph are evaluated to detect abnormal users. Firstly, the multi-view-angle judgment module adopts a view-level comparison method based on a complete graph, so that information in an attribute network can be comprehensively displayed; then, the graph reconstruction module fully integrates attributes and topological information; finally, wide experiments are performed on six common graph data sets. The method also adopts an AUC index to measure the anomaly detection performance. According to the method, the comprehensiveness of data reconstruction can be effectively considered, the unreliability of neighborhood information is reduced, the complexity is not increased, and compared with other methods, the method has better performance on financial data anomaly detection tasks and has certain practical value.

Description

Technical Field

[0001] The present invention belongs to the field of financial data anomaly detection, and in particular relates to a financial data anomaly detection method based on multi-channel graph contrast learning. Background Art

[0002] With the rapid development of information technology, the financial field has generated complex, interdependent and interrelated data, which are often represented in the form of graphs, known as attribute networks or attribute graphs. In financial attribute networks, nodes and edges represent user information and their complex interactive information. Among them, graph anomaly detection, as an important field, has practical applications not only in financial fraud detection, but also in social spam detection, telecommunications fraud detection and many other fields. Graph anomaly detection aims to identify patterns in attribute networks that deviate from the norms of most nodes. According to previous studies, anomalies can be divided into two types: attribute anomalies and topological anomalies. Attribute anomalies occur when node attributes are contaminated by factors such as noise, while topological anomalies arise due to incorrect links between nodes. Although the goal of anomaly detection is clear, anomaly detection in graphs faces major challenges due to the complexity of graph data, the lack of labeling information for real-world anomalies, and the diversity of anomaly patterns.

[0003] In response to these challenges, the field of anomaly detection in attribute networks has been extensively studied. Although early shallow anomaly detection techniques have been widely used, they cannot effectively capture deep nonlinear information due to their reliance on domain-specific knowledge. Fortunately, with the help of recent advances in deep learning technology, mainstream deep anomaly detection algorithms based on reconstruction-based learning and contrast-based learning have achieved impressive results. In order to make full use of diverse supervisory signals, these two paradigms are combined to form a hybrid learning paradigm. Despite the success of the hybrid learning paradigm, it still faces problems with unreliable neighborhood information and neglect of topological details. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the prior art and provide a financial data anomaly detection method based on multi-channel graph contrast learning, which can effectively have better performance than other methods in the graph anomaly detection task and has certain practical value.

[0005] To achieve the above object, the technical solution of the present invention is: a financial data anomaly detection method based on multi-channel graph contrast learning, comprising the following steps:

[0006] Step S1: Input financial data information documents, and obtain financial graph data by modeling user personal and interactive information, namely, feature matrix and adjacency matrix A. raw , generate the front view X pos and negative view Xneg , initialize the relevant parameters and initial values ​​of each problem;

[0007] Step S2: Using the original view X raw , Front View X pos and negative view X neg , respectively, through the graph convolutional network to perform multi-view level discrimination tasks and data reconstruction tasks, and calculate each node v i The discriminant anomaly score s dis,i and reconstruct anomaly score s rec,i , get the comprehensive anomaly score s of each node i ;

[0008] Step S3: For the multi-view level discrimination task, the binary cross entropy loss function is applied to calculate the loss function value; for the data reconstruction task, the mean square error loss function is applied; through continuous iteration, the prediction value and network parameters are updated, and if the loss value does not converge to a stable value, return to step S2;

[0009] Step S4: Output optimization to obtain the final anomaly score of each node.

[0010] In one embodiment of the present invention, the step S1 specifically includes the following steps:

[0011] Step S11: Use a public graph dataset to perform data preprocessing and complete label extraction;

[0012] Step S12: For the positive view, the original view is randomly masked, and the attributes of the masked nodes are set to zero; for the negative view, a random row transformation is applied to the node features in the original view, and the same random transformation is used as for the positive view. Figure 1 consistent masking strategy.

[0013] In one embodiment of the present invention, step S2 specifically includes the following steps:

[0014] Step S21: Encode the three views by using a two-layer graph convolutional network to obtain the node embedding matrix of the original view, the positive view and the negative view. The two-layer graph convolutional network is composed of GCN enc (·) represents that they are represented by the weight vector W enc Determine, the formula is as follows:

[0015] Z raw =GCN enc (X raw ,A;W enc )

[0016] Z pos =GCN enc (X pos ,A;W enc )

[0017] Z neg =GCN enc (X neg ,A;W enc )

[0018] Among them, Z raw , Z pos and Z neg The node embedding matrices representing the original view, positive view, and negative view respectively; the representation of each graph convolutional layer is as follows:

[0019]

[0020] Among them, H (l) and H (l+1) are the input and output of the lth convolutional layer respectively; represents the symmetric normalization of the adjacency matrix, I is the identity matrix, for The degree matrix of (l) represents the shared weight matrix of all nodes in the network; σ(·) represents the nonlinear activation function, using the ReLU activation function;

[0021] Step S22: For the multi-view level discrimination task, we use the bilinear layer Bilinear(·) to calculate the similarity between the original view and the positive view or negative view as the discrimination anomaly score, expressed as:

[0022] S pos =Bilinear(Z raw ,Z pos ; W dis )

[0023] S neg =Bilinear(Z raw ,Z neg ; W dis )

[0024] Among them, S pos represents the similarity between the original view and the front view, S neg represents the similarity between the original view and the negative view, W dis Represents the weight matrix of the discriminator; each node v i The calculation of the discriminant anomaly score is expressed as follows:

[0025] s dis,i =s neg,i -s pos,i

[0026] Among them, s pos,i Indicates v iThe front view score, s neg,i Indicates v i The negative view score of .

[0027] Step S23: For the data reconstruction task, use the first layer of graph convolutional network as the topology decoder GCN Tdec (·), the second layer graph convolutional network as attribute decoder GCN Adec (·), the reconstruction loss of the original view is calculated as the reconstruction anomaly score, expressed as:

[0028] H=GCN Tdec (Z raw ,A;W Tdec )

[0029]

[0030] Among them, H, W Tdec denote the latent node embedding matrix after the decoder and the weight matrix of the topological decoder respectively; represents the reconstructed adjacency matrix obtained by the inner product of H; W Adec Represents the weight matrix of the reconstructed attribute matrix and attribute decoder; each node v i The reconstruction anomaly score calculation is expressed as follows:

[0031]

[0032] s rec,i =(1-α)·s top,i +α·s att,i

[0033] Among them, a i and Represent A and The i-th row vector, x i and Represents X raw and The i-th row vector of s top,i Indicates v i The topological reconstruction score, s att,i Indicates v i The attribute reconstruction score, s rec,i represents the total reconstruction anomaly score, and α represents the balance parameter;

[0034] Step S24: The final comprehensive anomaly score is obtained by averaging through R rounds of calculations, which is expressed as:

[0035] s i =(1-β)·s dis,i +β·s rec,i

[0036]

[0037] Among them, β is used to balance the scores of the two modules, score(v i ) indicates v i The anomaly score, It represents the anomaly score of the rth round of testing, where R is the number of testing rounds.

[0038] In one embodiment of the present invention, step S3 specifically includes the following steps:

[0039] Step S31, obtain a local density loss function by applying conditions to the global density loss function, the formula of the local density loss function is as follows:

[0040]

[0041] Where N represents the number of nodes, y i The labels 1 and 0 represent the positive and negative views respectively, s dis,i Represents the original view and each node v i The similarity score between the generated views;

[0042] Step S32: Calculate the loss function value by the gradient of the local density loss function, and obtain the minimum loss value through continuous iteration. The formula is as follows:

[0043]

[0044] Among them, α is a parameter used to balance topology and attribute reconstruction;

[0045] The overall loss function formula is as follows:

[0046] Loss = (1-β) Loss dis +β·Loss rec

[0047] Step S33, iteratively update the network parameter W enc , W Tdec , W Adec , W dis ;

[0048] Step S34: If the loss function converges to a stable value, proceed to step S4; otherwise, return to step S2.

[0049] The present invention also provides a financial data anomaly detection system based on multi-channel graph contrast learning, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0050] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] The graph reconstruction module adopted by the present invention fully incorporates attribute and topological information, enhancing the comprehensiveness of data reconstruction learning. In addition, the multi-view discrimination module adopts a view-level comparison method based on the entire graph, which helps to fully present the information in the belonging network and alleviate the problem of unreliable neighborhood without increasing the complexity. Experiments show that the present invention is superior to other existing models and can have better performance than other methods in the task of graph anomaly detection, and has certain practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a flow chart of an embodiment of the present invention.

[0054] Figure 2 This is a diagram of the algorithm architecture of the present invention. DETAILED DESCRIPTION

[0055] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings.

[0056] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0057] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0058] like Figure 1 , 2 As shown, this embodiment provides a financial data anomaly detection method based on multi-channel graph contrast learning, comprising the following steps:

[0059] Step S1: Input financial data information documents, and obtain financial graph data by modeling user personal and interactive information, namely, feature matrix and adjacency matrix A. Use the input feature matrix, namely, original view X raw, generate the front view X pos and negative view X neg , initialize the relevant parameters and initial values ​​of each problem;

[0060] Step S2: Through these three views, the multi-view level discrimination task and data reconstruction task are performed through the graph convolutional network to calculate each node v i Discrimination anomaly score dis,i and reconstruct anomaly score s rec,i , get the comprehensive anomaly score s of each node i ;

[0061] Step S3: For multi-view discrimination tasks, the binary cross entropy loss function is applied to calculate the loss function value; for data reconstruction tasks, the mean square error loss function is applied. Through continuous iterations, the prediction value and network parameters are updated. If the loss value does not converge to a stable value, return to step S2;

[0062] Step S4: Output optimization to obtain the final anomaly score of each node.

[0063] Furthermore, the step S1 specifically includes the following steps:

[0064] Step S11: Use a public graph dataset to perform data preprocessing to complete label extraction;

[0065] Step S12: For the specific method of generating views, the positive view is randomly masked and the attributes of the masked nodes are set to zero. For the negative view, we apply random row transformations to the node features in the original view and use the same Figure 1 consistent masking strategy.

[0066] Furthermore, the step S2 specifically includes the following steps:

[0067] Step S21: The node embedding matrices of the original view, the positive view, and the negative view are first obtained by encoding the three views using a two-layer graph convolutional network. The two-layer graph convolutional network is composed of GCN. enc (·) indicates that they are mainly composed of the weight vector W enc Determine, the formula is as follows:

[0068] Z raw =GCN enc (X raw ,A;W enc )

[0069] Z pos =GCN enc (X pos ,A;W enc )

[0070] Zneg =GCN enc (X neg ,A;W enc )

[0071] Among them, Z raw , Z pos and Z neg The node embedding matrices representing the original view, positive view, and negative view respectively; the representation of each graph convolutional layer is as follows:

[0072]

[0073] Among them, H (l) and H (l+1) are the input and output of the lth convolutional layer respectively. represents the symmetric normalization of the adjacency matrix, I is the identity matrix, for The degree matrix of (l) Represents the shared weight matrix of all nodes in the network. σ(·) represents the nonlinear activation function, and the ReLU activation function is used here.

[0074] Step S22: We use the bilinear layer Bilinear(·) to calculate the similarity between the original view and the positive view or negative view as the discriminant anomaly score, expressed as:

[0075] S pos =Bilinear(Z raw ,Z pos ; W dis ),

[0076] S neg =Bilinear(Z raw ,Z neg ; W dis ).

[0077] Among them, S pos represents the similarity between the original view and the front view, S neg represents the similarity between the original view and the negative view, W dis Represents the weight matrix of the discriminator. Each node v i The calculation of the discriminant anomaly score is expressed as follows:

[0078] s ds,i =s neg,i -s dis,i

[0079] Among them, s pos,i Indicates v i The front view score, s neg,i Indicates vi The negative view score of .

[0080] Step S23: For the data reconstruction task, we use the first layer of graph convolutional network as the topology decoder GCN Tdec (·), the second layer graph convolutional network as attribute decoder GCN Adec (·),,the reconstruction loss of the original view is calculated as the reconstruction anomaly score, which can be expressed as:

[0081] H=GCN Tdec (Z raw ,A;W Tdec ),

[0082]

[0083] Among them, H, W Tdec Represents the latent node embedding matrix after the decoder and the weight matrix of the topological decoder. represents the reconstructed adjacency matrix obtained by the inner product of H. W Adec Represents the weight matrix of the reconstructed attribute matrix and attribute decoder. Each node v i The reconstruction anomaly score calculation is expressed as follows:

[0084]

[0085] s rec,i =(1-α)·s top,i +α·s att,i

[0086] Among them, a i and Represent A and The i-th row vector, x i and Represents X raw and The i-th row vector, s top,i Indicates v i The topological reconstruction score, s att,i Indicates v i The attribute reconstruction score, s rec,i represents the total reconstruction anomaly score, and α represents the balance parameter.

[0087] Step S24: The final comprehensive anomaly score is obtained by averaging through R rounds of calculations, which can be expressed as:

[0088] s i =(1-β)·s dis,i +β·s rec,i

[0089]

[0090] Among them, β is used to balance the scores of the two modules, score(v i ) indicates v i The anomaly score, represents the anomaly score of each test round, and R is the number of test rounds.

[0091] Furthermore, the step S3 specifically includes the following steps:

[0092] Step S31: A local density loss function is obtained by applying conditions to the global density loss function. The formula of the local density loss function is as follows:

[0093]

[0094] Where N represents the number of nodes, y i The labels 1 and 0 represent the positive and negative views respectively, s dis,i Represents the original view and each node v i The similarity score between the generated views.

[0095] Step S32: Calculate the loss function value through the gradient of the local density loss function, and obtain the minimum loss value through continuous iteration. The formula is as follows:

[0096]

[0097] Here, α is a parameter used to balance topology and attribute reconstruction.

[0098] The overall loss function formula is as follows:

[0099] Loss = (1-β) Loss dis +β·Loss rec

[0100] Step S33: Iteratively update the network parameters W enc , W Tdec , W Adec , W dis .

[0101] Step S34: If the loss function converges to a stable value, proceed to step S4, otherwise return to continue executing step S2.

[0102] This embodiment starts from practical application. First, a graph data set is obtained by modeling financial data documents, that is, the feature matrix X raw and adjacency matrix A, using the feature matrix, i.e., the original view X raw Generate Front View X pos and negative view X negThen, through these three views, the multi-view level discrimination task and data reconstruction task are performed through the graph convolutional network to calculate each node v i Discrimination anomaly score dis,i and reconstruct anomaly score s rec,i , get the comprehensive anomaly score s of each node i ; Then, for the multi-view discrimination task, the binary cross entropy loss function is applied to calculate the loss function value; for the data reconstruction task, the mean square error loss function is applied, and the prediction value and network parameters are updated through continuous iteration until the loss function converges; finally, the final anomaly score of each node obtained by optimization is output. This embodiment is based on multi-view discrimination and data reconstruction, effectively detects anomalies in the graph, and has certain application value.

[0103] The present invention also provides a financial data anomaly detection system based on multi-channel graph contrast learning, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0104] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0105] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0106] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0107] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0108] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0109] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A financial data anomaly detection method based on multi-channel graph contrast learning, characterized in that: The steps include: Step S1: Input financial data information documents, and obtain financial graph data by modeling user personal and interactive information, namely, feature matrix and adjacency matrix A. raw , generate the front view X pos and negative view X neg , initialize the relevant parameters and initial values ​​of each problem; Step S2: Using the original view X raw , Front View X pos and negative view X neg , respectively, through the graph convolutional network to perform multi-view level discrimination tasks and data reconstruction tasks, and calculate each node v i The discriminant anomaly score s dis,i and reconstruct anomaly score s rec,i , get the comprehensive anomaly score s of each node i ; Step S3: For the multi-view level discrimination task, the binary cross entropy loss function is applied to calculate the loss function value; for the data reconstruction task, the mean square error loss function is applied; through continuous iteration, the prediction value and network parameters are updated, and if the loss value does not converge to a stable value, return to step S2; Step S4: Output optimization to obtain the final anomaly score of each node.

2. The method for detecting anomalies in financial data based on multi-channel graph contrast learning according to claim 1, characterized in that: The step S1 specifically includes the following steps: Step S11: Use a public graph dataset to perform data preprocessing and complete label extraction; Step S12: For the front view, the original view is randomly masked, and the attributes of the masked nodes are set to zero; for the negative view, random row transformation is applied to the node features in the original view, and a masking strategy consistent with the front view is used.

3. The method for detecting anomalies in financial data based on multi-channel graph contrast learning according to claim 1, characterized in that: The step S2 specifically includes the following steps: Step S21: Encode the three views by using a two-layer graph convolutional network to obtain the node embedding matrix of the original view, the positive view and the negative view. The two-layer graph convolutional network is composed of GCN enc (·) represents that they are represented by the weight vector W enc Determine, the formula is as follows: Z raw =GCN enc (X raw ,A;W enc ) Z pos =GCN enc (X pos ,A;W enc ) Z neg =GCN enc (X neg ,A;W enc ) Among them, Z raw , Z pos and Z neg The node embedding matrices representing the original view, positive view, and negative view respectively; the representation of each graph convolutional layer is as follows: Among them, H (l) and H (l+1) are the input and output of the lth convolutional layer respectively; represents the symmetric normalization of the adjacency matrix, I is the identity matrix, for The degree matrix of (l) represents the shared weight matrix of all nodes in the network; σ(·) represents the nonlinear activation function; Step S22: For the multi-view level discrimination task, the similarity between the original view and the positive view or the negative view is calculated as the discrimination anomaly score; Step S23: For the data reconstruction task, calculate the reconstruction loss of the original view as the reconstruction anomaly score; Step S24: averaging the results through R rounds of calculations to obtain the final comprehensive anomaly score.

4. The method for detecting anomalies in financial data based on multi-channel graph contrast learning according to claim 3, characterized in that: σ(·) uses the ReLU activation function.

5. According to the method for detecting anomalies in financial data based on multi-channel graph contrast learning in claim 3, a bilinear layer Bilinear(·) is used to calculate the similarity between the original view and the positive view or the negative view as the discriminant anomaly score, which is expressed as: S pos =Bilinear(Z raw ,WITH pos ;IN dis ) S neg =Bilinear(Z raw ,WITH neg ;IN dis ) in, S pos represents the similarity between the original view and the front view, S neg represents the similarity between the original view and the negative view, W dis Represents the weight matrix of the discriminator; each node v i The calculation of the discriminant anomaly score is expressed as follows: s dis,i =s neg,i -s pos,i Among them, s pos,i Indicates v i The front view score, s neg,i Indicates v i The negative view score of .

6. The method for detecting anomalies in financial data based on multi-channel graph contrast learning according to claim 5, characterized in that: Reconstructing the anomaly score, using the first layer of the graph convolutional network as the topological decoder GCN Tdec (·), the second layer graph convolutional network as attribute decoder GCN Adec (·), specifically expressed as: It is expressed as: H=GCN Tdec (Z raw ,A;W Tdec ) Among them, H, W Tdec denote the latent node embedding matrix after the decoder and the weight matrix of the topological decoder respectively; represents the reconstructed adjacency matrix obtained by the inner product of H; W Adec Represents the weight matrix of the reconstructed attribute matrix and attribute decoder; each node v i The reconstruction anomaly score calculation is expressed as follows: s rec,i =(1-α)·s top,i +a·s att,i Among them, a i and Represent A and The i-th row vector, x i and Represents X raw and The i-th row vector of s top,i Indicates v i The topological reconstruction score, s att,i Indicates v i The attribute reconstruction score, s rec,i represents the total reconstruction anomaly score, and α represents the balance parameter.

7. The method for detecting anomalies in financial data based on multi-channel graph contrast learning according to claim 6, characterized in that: The comprehensive anomaly score is expressed as: s i =(1-β)·s dis,i +β·s rec,i Among them, β is used to balance the scores of the two modules, score(v i ) indicates v i The anomaly score, It represents the anomaly score of the rth round of testing, where R is the number of testing rounds.

8. The method for detecting anomalies in financial data based on multi-channel graph contrast learning according to claim 3, characterized in that: The step S3 specifically comprises the following steps: Step S31, obtain a local density loss function by applying conditions to the global density loss function, the formula of the local density loss function is as follows: Where N represents the number of nodes, y i The labels 1 and 0 represent the positive and negative views respectively, s dis,i Represents the original view and each node v i The similarity score between the generated views; Step S32: Calculate the loss function value by the gradient of the local density loss function, and obtain the minimum loss value through continuous iteration. The formula is as follows: Among them, α is a parameter used to balance topology and attribute reconstruction; The overall loss function formula is as follows: Loss=(1-β)·Loss dis +β·Loss rec Step S33, iteratively update the network parameter W enc , W Tdec , W Adec , W dis ; Step S34: If the loss function converges to a stable value, proceed to step S4; otherwise, return to step S2.

9. A financial data anomaly detection system based on multi-channel graph contrast learning, characterized in that: The method comprises a memory, a processor and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps as claimed in any one of claims 1 to 8 can be implemented.

10. A computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor, and when the processor executes the computer program instructions, the method steps according to any one of claims 1 to 8 can be implemented.

Images