A method and related equipment for enterprise risk assessment based on dynamic graph neural network

Abstract

The present application relates to the technical field of enterprise risk assessment, and provides an enterprise risk assessment method based on a dynamic graph neural network and related equipment. The method includes: obtaining the relationship data of multiple target enterprises; constructing a multi-layer enterprise social network based on all the relationship data; extracting features from the multi-layer enterprise social network using a dynamic graph neural network to obtain short-term feature embedding, long-term feature embedding, and topological feature embedding of the multi-layer enterprise social network; obtaining persistent features based on the multi-layer enterprise social network, and fusing the persistent features, short-term feature embedding, long-term feature embedding, and topological feature embedding to obtain fused features; performing risk assessment on each target enterprise based on the fused features and the multi-layer enterprise social network to obtain the risk assessment results of each target enterprise. The method of the present application can improve the accuracy of enterprise risk assessment.

Description

Technical Field

[0001] The present application relates to the technical field of enterprise risk assessment, and in particular to an enterprise risk assessment method and related equipment based on a dynamic graph neural network. Background Art

[0002] In the current globalized and digitalized business environment, enterprises have formed a complex multi-level relationship network, covering multi-dimensional connections such as supply chain dependence, technical cooperation, and capital association. In the field of enterprise risk management, existing enterprise intelligence systems (such as Tianyancha, Qichacha, etc.) mainly use graph databases to store enterprise entity relationships and conduct risk assessments through preset risk identification rules and traditional graph algorithms. For example, typical risk transmission analysis uses static graph algorithms such as the shortest path or PageRank to score and sort the relationships between enterprises. Although enterprise association analysis and risk monitoring based on knowledge graphs have been realized, which can provide a visual display of inter-enterprise relationships and basic risk warnings, this method based on deterministic rules and static graph analysis has obvious limitations: first, it cannot effectively model the temporal dynamic characteristics of enterprise relationships, ignoring the accumulation and evolution effects of risks in the time dimension; second, it simplifies multi-dimensional enterprise relationships (such as equity control, supply chain dependence, technical cooperation, etc.) into a single plane network, resulting in inaccurate identification of risk transmission paths; third, there is a lack of in-depth analysis of the network topology structure, making it difficult to discover systemic risks hidden in complex relationship structures.

[0003] In view of the above challenges, existing technical solutions have not yet provided a comprehensive solution, especially in the mining of dynamic evolution characteristics of enterprise relationship networks, the integration of local network structures and global topological characteristics, and the modeling of interactions between multi-layer networks. As a result, the current enterprise risk assessment methods have the problem of low accuracy in enterprise risk assessment. Summary of the invention

[0004] The present application provides an enterprise risk assessment method and related equipment based on a dynamic graph neural network, which can solve the problem of low accuracy in enterprise risk assessment.

[0005] In a first aspect, the present application provides an enterprise risk assessment method based on a dynamic graph neural network, the enterprise risk assessment method comprising:

[0006] Acquire relationship data of multiple target enterprises; the relationship data of each target enterprise is used to describe the enterprise relationship between the target enterprise and other target enterprises, and the enterprise relationship is one of the multiple relationships;

[0007] Build a multi-layer enterprise social network based on all relationship data; the multi-layer enterprise social network corresponds to multiple relationships one by one, and multiple nodes in the multi-layer enterprise social network correspond to multiple target enterprises one by one. The edge between two nodes in the same layer is the enterprise relationship between the corresponding two target enterprises, and the edge between two nodes in different layers is the influence relationship between the corresponding two target enterprises.

[0008] The dynamic graph neural network is used to extract features from multi-layer enterprise social networks, and the short-term feature embedding, long-term feature embedding and topological feature embedding of the multi-layer enterprise social network are obtained; the short-term feature embedding is used to describe the enterprise relationship information in the multi-layer enterprise social network in a short time span, the long-term feature embedding is used to describe the enterprise relationship information in the multi-layer enterprise social network in a long time span, and the topological feature embedding is used to describe the topological structure information of the multi-layer enterprise social network;

[0009] Based on the multi-layer enterprise social network, persistent features are obtained, and the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding are fused to obtain fused features; the persistent features are used to describe the stability of each edge in the multi-layer enterprise social network;

[0010] A risk assessment is performed on each target enterprise based on the fusion characteristics and the multi-layer enterprise social network to obtain a risk assessment result for each target enterprise; the risk assessment result is used to describe the risk level of the target enterprise.

[0011] Optionally, build a multi-layer enterprise social network based on all relationship data, including:

[0012] For each relationship, target enterprises with relationships in the relationship data of all target enterprises are used as hierarchical target enterprises of the relationship, and a single-layer enterprise social network of the relationship is generated according to the relationship data of all hierarchical target enterprises;

[0013] The influence relationship between every two different single-layer enterprise social networks is calculated, and the edges between the nodes in different single-layer enterprise social networks are generated based on all the influence relationships to obtain a multi-layer enterprise social network.

[0014] Optionally, the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding are fused to obtain fused features, including:

[0015] Embed the short-term features and fuse the persistent features to obtain the first feature;

[0016] The first feature and the long-term feature embedding are integrated to obtain the second feature;

[0017] The second feature and the topological feature are embedded and integrated to obtain the fused feature.

[0018] Optionally, the short-term feature embedding and the persistent feature are fused to obtain a first feature, including:

[0019] By formula:

[0020]

[0021]

[0022]

[0023] Calculate the first feature ;

[0024] in, represents the dynamic attention weight, Indicates The first node feature of the node, represents the first The short-term characteristics of nodes, Represents a persistent feature, represents the scoring function, represents the first The short-term characteristics of nodes, , Representation Node The neighbor nodes of Represents a collection of nodes in a multi-layer enterprise social network.

[0025] Optionally, the first feature and the long-term feature embedding are fused to obtain a second feature, including:

[0026] By formula:

[0027]

[0028]

[0029]

[0030] Calculate the second feature ;

[0031] in, Indicates The second node feature of the node, Represents the first The node and The weights between nodes, Represents the first The neighboring nodes of a node, represents the enhanced long-term feature embedding, represents the first The components corresponding to the nodes, represents the first The components corresponding to the nodes;

[0032] The second feature and the topological feature are embedded and integrated to obtain the fusion feature, including:

[0033] By formula:

[0034]

[0035]

[0036] Calculate fusion features ;

[0037] in, represents a graph neural network, represents the weight, represents the enhanced topological features, Indicates the topological features The components corresponding to the nodes, Indicates the topological features The component corresponding to each node.

[0038] Optionally, risk assessment is performed on each target enterprise based on the fusion features and the multi-layer enterprise social network to obtain risk assessment results for each target enterprise, including:

[0039] For each target enterprise, carry out the following steps:

[0040] Calculate the centrality scores of the nodes corresponding to the target enterprise based on the multi-layer enterprise social network;

[0041] Based on the centrality score and fusion characteristics, the risk assessment of the target enterprise is carried out to obtain the risk assessment result of the target enterprise.

[0042] Optionally, after the step of obtaining persistent features based on the multi-layer enterprise social network and fusing the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features, the enterprise risk assessment method further includes:

[0043] For each target enterprise, a comprehensive opportunity score is calculated for each target enterprise based on multi-layer enterprise social networks and persistence characteristics; the comprehensive opportunity score is used to describe the probability that the target enterprise has potential development opportunities.

[0044] In a second aspect, the present application provides an enterprise risk assessment device based on a dynamic graph neural network, comprising:

[0045] An acquisition module is used to acquire relationship data of multiple target enterprises; the relationship data of each target enterprise is used to describe the enterprise relationship between the target enterprise and other target enterprises, and the enterprise relationship is one of the multiple relationships;

[0046] A construction module is used to construct a multi-layer enterprise social network based on all relationship data; the multi-layer enterprise social network corresponds one-to-one to multiple relationships, multiple nodes in the multi-layer enterprise social network correspond one-to-one to multiple target enterprises, the edge between two nodes in the same layer is the enterprise relationship between the corresponding two target enterprises, and the edge between two nodes in different layers is the influence relationship between the corresponding two target enterprises;

[0047] A feature extraction module is used to extract features from a multi-layer enterprise social network using a dynamic graph neural network to obtain short-term feature embedding, long-term feature embedding and topological feature embedding of the multi-layer enterprise social network; short-term feature embedding is used to describe enterprise relationship information in a multi-layer enterprise social network over a short time span, long-term feature embedding is used to describe enterprise relationship information in a multi-layer enterprise social network over a long time span, and topological feature embedding is used to describe topological structure information of a multi-layer enterprise social network;

[0048] A fusion module is used to obtain persistent features based on multi-layer enterprise social networks, and fuse persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features; persistent features are used to describe the stability of each edge in the multi-layer enterprise social network;

[0049] The risk assessment module is used to conduct risk assessment on each target enterprise based on the fusion characteristics and multi-layer enterprise social network to obtain the risk assessment results of each target enterprise; the risk assessment results are used to describe the risk level of the target enterprise.

[0050] In a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned enterprise risk assessment method based on dynamic graph neural network when executing the above-mentioned computer program.

[0051] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned enterprise risk assessment method based on dynamic graph neural network.

[0052] The above solution of the present application has the following beneficial effects:

[0053] In an embodiment of the present application, by acquiring the relationship data of multiple target enterprises, and then constructing a multi-layer enterprise social network based on all the relationship data, and then using a dynamic graph neural network to extract features from the multi-layer enterprise social network, short-term feature embedding, long-term feature embedding and topological feature embedding of the multi-layer enterprise social network are obtained, and then persistent features are obtained based on the multi-layer enterprise social network, and the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding are fused to obtain fused features, and finally, risk assessment is performed on each target enterprise based on the fused features and the multi-layer enterprise social network to obtain the risk assessment results of each target enterprise. Among them, constructing a multi-layer enterprise social network can fully describe the association relationship between target enterprises, realize accurate modeling of enterprise relationships, fuse persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features, improve the information richness and comprehensiveness of fused features, and perform risk assessment based on information-rich fused features and accurate multi-layer enterprise social networks, which can effectively improve the accuracy of enterprise risk assessment.

[0054] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 A flowchart of a method for enterprise risk assessment based on a dynamic graph neural network provided in one embodiment of the present application;

[0057] Figure 2 A schematic diagram of the structure of a multi-layer enterprise social network provided in one embodiment of the present application;

[0058] Figure 3 A schematic diagram of the structure of an enterprise risk assessment device based on a dynamic graph neural network provided in one embodiment of the present application;

[0059] Figure 4 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0060] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0061] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0062] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0063] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0064] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0065] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0066] In view of the low accuracy of existing enterprise risk assessment, the embodiment of the present application provides an enterprise risk assessment method based on a dynamic graph neural network. The enterprise risk assessment method obtains the relationship data of multiple target enterprises, then constructs a multi-layer enterprise social network based on all the relationship data, and then uses a dynamic graph neural network to extract features of the multi-layer enterprise social network to obtain short-term feature embedding, long-term feature embedding and topological feature embedding of the multi-layer enterprise social network. Then, based on the multi-layer enterprise social network, persistent features are obtained, and the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding are fused to obtain fused features. Finally, risk assessment is performed on each target enterprise based on the fused features and the multi-layer enterprise social network to obtain the risk assessment results of each target enterprise. Among them, constructing a multi-layer enterprise social network can fully describe the association relationship between target enterprises, realize accurate modeling of enterprise relationships, fuse persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features, improve the information richness and comprehensiveness of fused features, and perform risk assessment based on information-rich fused features and accurate multi-layer enterprise social networks, which can effectively improve the accuracy of enterprise risk assessment.

[0067] Next, an exemplary description is given of the enterprise risk assessment method based on dynamic graph neural network provided in this application.

[0068] like Figure 1 As shown, the enterprise risk assessment method based on dynamic graph neural network provided by this application includes the following steps:

[0069] Step 11: Obtain relationship data of multiple target enterprises.

[0070] The relationship data of each target enterprise is used to describe the enterprise relationship between the target enterprise and other target enterprises (the relationship data includes the enterprise relationship between the target enterprise and at least one other target enterprise). The enterprise relationship (such as project cooperation, results sharing, market competition, supply chain relationship, parent-subsidiary company, the same director, the same shareholder, etc.) belongs to one of the multiple relationships (such as cooperative relationship, supply relationship, competitive relationship, subsidiary relationship, and associated relationship). For example, the enterprise relationship: project cooperation and results sharing belong to cooperative relationship, supply chain relationship belongs to supply relationship, market competition belongs to competitive relationship, parent-subsidiary formula belongs to subsidiary relationship, and the same director and the same shareholder belong to associated relationship.

[0071] It should be noted that the above relationship data also includes the time information of the enterprise relationship between the target enterprises, which can describe the generation time and end time of the relationship between the enterprises. For example, for Enterprise A, the relationship data of Enterprise A is: from 2016 to 2017, it had a cooperative relationship with Enterprise B, and from 2018 to date, it has a subsidiary relationship with Enterprise B.

[0072] In some embodiments of the present application, the relationship data of the target enterprise can be obtained by visiting websites that disclose enterprise relationship data, such as Qichacha and Tianyancha.

[0073] Step 12: construct a multi-layer enterprise social network based on all relationship data.

[0074] The above multi-layer enterprise social network corresponds to multiple relationships one by one, and multiple nodes in the multi-layer enterprise social network correspond to multiple target enterprises one by one. The edge between two nodes in the same layer is the enterprise relationship between the corresponding two target enterprises, and the edge between two nodes in different layers is the influence relationship between the corresponding two target enterprises (used to describe the mutual influence between the enterprise relationship of a node in a layer and the enterprise relationship of a node in another layer, such as supply chain relationships often promote cooperative relationships, and competitive relationships often hinder cooperative relationships). Each node has a set of characteristics of the corresponding target enterprise, including basic characteristics (enterprise name, industry type, number of employees, location), financial status (revenue, profit, total assets), operating status (recent financing, credit rating, market share), etc.

[0075] In an embodiment of the present application, the above-mentioned step of constructing a multi-layer enterprise social network based on all relationship data includes:

[0076] In the first step, for each relationship, the target enterprises with relationships in the relationship data of all target enterprises are taken as the hierarchical target enterprises of the relationship, and a single-layer enterprise social network of the relationship is generated based on the relationship data of all hierarchical target enterprises.

[0077] Specifically, if there is an enterprise relationship corresponding to the relationship between two hierarchical target enterprises, an edge is generated.

[0078] It should be noted that the edges in a single-layer enterprise social network reflect the direct relationships between enterprises at the same level.

[0079] For example, for cooperative relationships, among enterprise A, enterprise B, enterprise C, and enterprise D, enterprise A has a cooperative relationship with enterprise B and enterprise C, and enterprise C has a cooperative relationship with enterprise D. Then, in a single-layer enterprise social network for cooperative relationships, there is an edge between the node corresponding to enterprise A and the node corresponding to enterprise B and the node corresponding to enterprise C, and there is an edge between the node corresponding to enterprise C and the node corresponding to enterprise D.

[0080] It should be noted that after the edge is generated, the attributes of each edge are attached with the timestamp in the corresponding relationship data to reflect the time information. At the same time, the attributes of the edge include the enterprise relationship and relationship strength (such as the number of cooperation times, dependence, etc.) of the corresponding two target enterprises.

[0081] In the second step, the influence relationship between every two different single-layer enterprise social networks is calculated, and the edges between the nodes in different single-layer enterprise social networks are generated based on all the influence relationships to obtain a multi-layer enterprise social network.

[0082] It should be noted that the edges between nodes in different single-layer enterprise social networks reflect the interweaving and mutual influence of relationships between different levels, revealing the interaction between layers. For example, the dependency relationship at the supply chain layer may affect the risk intensity of the cooperation layer.

[0083] Exemplarily, the influence relationship can be obtained by calculating indicators such as the similarity between single-layer enterprise social networks (which can be calculated using calculation formulas such as cosine similarity). In order to analyze the mutual influence between the single-layer enterprise social network of the cooperative relationship and the single-layer enterprise social network of the supply relationship, the similarity indicators of the two networks (such as cosine similarity) are first calculated. If the similarity is greater than the set threshold, it is considered that there is a significant influence relationship between the two networks. On this basis, cross-layer edges are further generated for the relevant nodes in the two networks based on the relationship strength, interaction frequency or time dependence. The weight of the cross-layer edge can be quantified according to the similarity value, time decay factor or interaction intensity to reflect the strength of the influence relationship between the two layers of networks. And generate a cross-layer correlation matrix ,in The cross-layer influence relationship and influence degree are quantified, indicating how the relationship between target enterprises vi and vj in the supply layer l affects the relationship in the cooperation layer m.

[0084] It should be noted that in a multi-layer enterprise social network, each target enterprise will only correspond to one node. For target enterprises with multiple relationships (which exist in multiple single-layer enterprise social networks), after forming a multi-layer enterprise social network, only one node will be retained, and the nodes in other layers are equivalent to the mapping of this node, which is convenient for the structural analysis and display of the multi-layer enterprise social network. For example, if enterprise A has a cooperative relationship with two target enterprises and a competitive relationship with one target enterprise, in the multi-layer enterprise social network, the node corresponding to enterprise A should have two edges with the attribute of cooperative relationship and one edge with the attribute of competitive relationship, instead of the node corresponding to enterprise A in the cooperative layer having two edges with the attribute of cooperative relationship, and the other node corresponding to enterprise A in the competitive layer having one edge with the attribute of cooperative relationship.

[0085] Since the edges between nodes in different layers have timestamps, when analyzing multi-layer enterprise social networks, the timeliness of the relationships in the multi-layer enterprise social networks is dynamically updated according to the timestamps. For example, if it is necessary to analyze the relationships of the target enterprises between 2015 and 2017, the edges with timestamps ending before 2015 and starting after 2017 are marked as invalid.

[0086] It is worth mentioning that building a multi-layer enterprise social network can fully describe the relationship between target enterprises, realize accurate modeling of enterprise relationships, and facilitate the time-effectiveness analysis of the relationships between enterprises.

[0087] The following is an illustrative description of a multi-layer enterprise social network with reference to a specific example.

[0088] Multi-layer enterprise social network Figure 2 As shown, it includes five layers of network, namely cooperation layer, supply layer, competition layer, subsidiary layer and association layer. A, B, C are the numbers of nodes. Solid lines represent the edges between nodes in the layer, and dotted lines represent the edges between nodes between layers.

[0089] Step 13, using a dynamic graph neural network to extract features of the multi-layer enterprise social network, and obtain short-term feature embedding, long-term feature embedding and topological feature embedding of the multi-layer enterprise social network.

[0090] The above short-term feature embedding is used to describe the enterprise relationship information in the multi-layer enterprise social network over a short time span, the long-term feature embedding is used to describe the enterprise relationship information in the multi-layer enterprise social network over a long time span, and the topological feature embedding is used to describe the topological structure information of the multi-layer enterprise social network. The short time span and the long time span are set according to the actual cycle requirements of enterprise risk assessment. The short time span is smaller than the long time span. For example, the short time span is one quarter, and the long time span can be one year.

[0091] It should be noted that the dynamic graph neural network is composed of a temporal graph sub-network (TGN), a dual graph attention sub-network (DGAT), and a graph convolutional neural network (GCN) connected in sequence. The input end of the temporal graph sub-network is the input end of the dynamic graph neural network, the output end of the temporal graph sub-network outputs short-term feature embedding, the output end of the dual graph attention sub-network outputs long-term feature embedding, and the output end of the graph convolutional neural sub-network outputs topological feature embedding.

[0092] Specifically, the multi-layer enterprise social network is input into the dynamic graph neural network to obtain the short-term feature embedding, long-term feature embedding and topological feature embedding of the multi-layer enterprise social network.

[0093] Step 14, obtaining persistent features based on the multi-layer enterprise social network, and fusing the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features.

[0094] The above persistence characteristics are used to describe the stability of each edge in the multi-layer enterprise social network.

[0095] In some embodiments of the present application, the step of obtaining persistent features based on a multi-layer enterprise social network, and fusing persistent features, short-term feature embedding, long-term feature embedding, and topological feature embedding to obtain fused features includes:

[0096] The first step is to obtain persistent features based on multi-layer enterprise social networks.

[0097] For example, the persistent homology method can be used to perform topological analysis on multi-layer enterprise social networks to obtain persistent barcodes, and then the persistent barcodes can be converted into persistent features using embedding algorithms, so that they can be embedded in the same feature space with other features for fusion, ensuring that the persistent features reflect their stability in the fusion.

[0098] The above-mentioned persistent homology method extracts topological structural features in multi-layer enterprise social networks, such as connected branches and loops, by constructing Rips complexes (used to describe the topological structure of connections between nodes in multi-layer enterprise social networks) under different thresholds to reveal potential risks and opportunities in the network. The selection of the distance threshold ϵ is usually based on the characteristics of the data and the analysis objectives, and is generally determined by node characteristics. For example, in the analysis of multi-layer enterprise social networks, ϵ can be set to the average strength of interactions between nodes, and the 75% or 90% quantile of the average strength can be taken as the initial threshold to ensure that the main structural features are captured. Experiments are conducted using different thresholds ϵ to find the threshold range in which the topological features are most stable under these thresholds. The Rips complex of the network is generated under the threshold ϵ. This step is repeated under different ϵ values ​​to capture the generation and disappearance of each topological structure (such as loops, holes, etc.) under different thresholds. At the same time, the duration of the topological structure is quantified by the persistence barcode. The length of the persistence barcode reflects the stability of the topological feature. Barcodes with high persistence represent long-term relationships, and barcodes with low persistence point to short-term or unstable relationships. It is used to analyze key relationship patterns in multi-layer enterprise social networks.

[0099] For each topological structure that is a hole (in a network, a hole can be regarded as a missing connection or structure, and their existence may indicate potential risks or opportunities. For example, in a supply relationship, a hole may indicate a potential supply relationship interruption point; in a social network, a hole may indicate a split or faction in a community), record its birth and death time under different thresholds ϵ, which can be automatically calculated by a persistent homology software package (such as Gudhi, Dionysus, etc.). The persistence measure is the difference between the death time and the birth time of the hole, expressed as , where di is the time of death of the hole and bi is the time of birth of the hole.

[0100] Persistence metrics can be used to identify risks and opportunities for vulnerabilities, e.g., to set risk thresholds 1.0, persistence metric values ​​above this value indicate high risk holes; opportunity threshold The value is 0.5, and values ​​below this value are potential opportunity holes.

[0101] In the second step, the short-term feature embedding and the persistent feature are fused to obtain the first feature.

[0102] Specifically, through the formula:

[0103]

[0104]

[0105]

[0106] Calculate the first feature ;

[0107] in, represents the dynamic attention weight, Indicates The first node feature of the node, represents the first The short-term characteristics of nodes, Represents a persistent feature, represents the scoring function, represents the first The short-term characteristics of nodes, , Representation Node The neighbor nodes of Represents a collection of nodes in a multi-layer enterprise social network.

[0108] It should be noted that in the above formula, an attention weight is defined , by calculating the attention weights of short-term feature embeddings and persistent features in each time step to capture the balance between temporal dynamic information and network structure stability. Short-term feature embeddings are used to capture the short-term dynamics of relationships, while persistent features reveal relatively stable long-term relationships in the network. At each time step, based on the dynamic attention weights, the scoring function of short-term feature embeddings and persistent features is calculated. , adaptively adjust the fusion weights to achieve the fusion of short-term and long-term features to cope with dynamic changes in the network.

[0109] Before this step, the persistent features, short-term feature embedding, long-term feature embedding, and topological feature embedding need to be time-step aligned and preprocessed, including:

[0110] (1) Time step alignment: First, arrange the persistent features, short-term feature embeddings, long-term feature embeddings, and topological feature embeddings by timestamp to determine a unified time scale. Secondly, use the same time scale for different features to ensure that all features are aligned at the same time step to capture potential risks in dynamic networks. Finally, use linear interpolation to handle missing values ​​to ensure that each time step has valid data.

[0111] (2) Adaptive time window selection: Adaptively adjust the time window of each feature according to the dynamic changes of enterprise relationships. For example, for short-term feature embedding, select a shorter time window (1 to 3 days) to capture rapid changes; for topological feature embedding and persistent features, select a longer time window (7 to 30 days) to reflect more stable structural features.

[0112] (3) Embedding normalization: First, use Z-score normalization to normalize the three feature embeddings separately. This ensures that the features are on the same scale to eliminate the impact of output differences between different models. The formula for Z-score normalization is:

[0113]

[0114] in, represents the standardized data, represents the data before normalization, represents the standard deviation, Represents the mean.

[0115] The third step is to fuse the first feature and the long-term feature embedding to obtain the second feature.

[0116] Specifically, through the formula:

[0117]

[0118]

[0119]

[0120] Calculate the second feature .

[0121] in, Indicates The second node feature of the node, Represents the first The node and The weights between nodes, Represents the first The neighboring nodes of a node, represents the enhanced long-term feature embedding, represents the first The components corresponding to the nodes, represents the first The component corresponding to each node.

[0122] It should be noted that based on the first feature, long-term feature embedding is used to further focus on the relationship between nodes and the interaction between layers. The first feature and the long-term feature embedding are weighted updated to ensure that the features of the inter-layer correlation are properly focused. For example, if there is a strong correlation between the supply layer and the cooperation layer, the cross-layer information of the correlation is prioritized through the attention mechanism in the long-term feature embedding.

[0123] The fourth step is to embed and integrate the second feature and the topological feature to obtain the fused feature.

[0124] Specifically, through the formula:

[0125]

[0126]

[0127] Calculate fusion features .

[0128] in, represents a graph neural network, represents the weight, represents the enhanced topological features, Indicates the topological features The components corresponding to the nodes, Indicates the topological features The component corresponding to each node.

[0129] It should be noted that in actual use, the dynamic changes of persistence characteristics are continuously monitored and the formula is used:

[0130]

[0131] The persistent features are updated, and when the method of the present application is used to perform enterprise risk assessment in the future, the persistent features updated in real time are used to participate in the fusion of this step to obtain fused features.

[0132] in, represents the updated persistent characteristic, represents the persistence feature at time step t, is a dynamic weight.

[0133] It is worth mentioning that by updating the persistent features in real time, the timeliness of the persistent features can be improved, and then the timeliness of the fused features can be improved. The persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding are fused to obtain the fused features, which improves the information richness and comprehensiveness of the fused features.

[0134] Step 15: Perform risk assessment on each target enterprise based on the fusion features and the multi-layer enterprise social network to obtain a risk assessment result for each target enterprise.

[0135] The above risk assessment results are used to describe the risk level of the target enterprise, and can be used by users to conduct risk analysis on their own or other enterprises, focus on and conduct manual analysis on enterprises with high risk levels, and analyze whether the enterprise currently has risks in supply chain, technological cooperation, capital connection, market competition, legal compliance, market operation and systemic risks, and then take corresponding measures or avoid cooperation with enterprises with high risk levels.

[0136] In some embodiments of the present application, the step of performing risk assessment on each target enterprise based on the fusion feature and the multi-layer enterprise social network to obtain the risk assessment result of each target enterprise includes:

[0137] For each target enterprise, carry out the following steps:

[0138] In the first step, the centrality scores of the nodes corresponding to the target enterprise are calculated based on the multi-layer enterprise social network.

[0139] Exemplarily, the above centrality score may be the betweenness centrality or degree centrality of the node in the multi-layer enterprise social network.

[0140] In the second step, risk assessment of the target enterprise is conducted based on the centrality score and fusion characteristics to obtain the risk assessment results of the target enterprise.

[0141] For example, the comprehensive risk score of the target enterprise can be calculated by the centrality score and the fusion feature, and the risk level of the target enterprise can be determined based on the comprehensive risk score and the preset threshold. The second preset threshold is the boundary between low risk and medium risk. is the boundary between medium risk and high risk. If the comprehensive risk score < , it is defined as low risk. , it is defined as medium risk. , it is defined as high risk.

[0142] Specifically, it can be expressed by the formula:

[0143]

[0144] Calculate the Comprehensive risk score of target companies .

[0145] in, , , is the weight parameter, Indicates The centrality score of a node, represents the risk transmission intensity extracted from the cross-layer association matrix of multi-layer enterprise social networks, and The interaction risk of nodes in multi-layer enterprise social networks, represents the first The persistence measure of the high-risk barcodes corresponding to the nodes indicates the systemic risk in the topology.

[0146] It should be noted that the method of the present application can also identify potential opportunities for target enterprises, and identify target enterprises with high cooperation willingness or supply chain potential as opportunity enterprises. Specifically, for each target enterprise, a comprehensive opportunity score is calculated for each target enterprise based on multi-layer enterprise social networks and persistence characteristics; the comprehensive opportunity score is used to describe the probability that the target enterprise has potential development opportunities. The higher the comprehensive opportunity score, the greater the probability that the target enterprise has potential development opportunities. Users can pay attention to enterprises with high comprehensive opportunity scores and consider developing cooperative or supply relationships with them.

[0147] Specifically, it can be expressed by the formula:

[0148]

[0149] Calculate the Comprehensive opportunity score for each target company .

[0150] in, is the comprehensive opportunity score of node i, For the The connectivity of each node indicates its potential cooperation strength in different layers of the network. For the The cross-layer similarity of nodes (such as industry similarity, technology similarity) is calculated through the cross-layer association matrix. The first The persistence measure of the low-risk barcode corresponding to the node represents the stable relationship potential of the node. , , is a weight parameter that adjusts the contribution of connectivity, similarity, and topological features to the chance score.

[0151] It is worth mentioning that building a multi-layer enterprise social network can fully describe the relationship between target enterprises, realize the accurate modeling of enterprise relationships, fuse the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features, which improves the information richness and comprehensiveness of the fused features. Risk assessment based on the information-rich fused features and accurate multi-layer enterprise social networks can effectively improve the accuracy of enterprise risk assessment.

[0152] In addition, the main advantages of the method of the present application include:

[0153] (1) Combining dynamic graph neural networks with persistent coherence: By integrating persistent features with the embedding of dynamic graph neural networks, the ability to recognize complex relationships in multi-layer enterprise social networks is improved, effectively capturing the characteristics of temporal changes and topological structures.

[0154] (2) Adaptive feature alignment: Adaptive time window selection and feature normalization techniques are used to ensure the consistency of features in the time dimension and enhance adaptability and accuracy.

[0155] (3) Dynamic attention mechanism: Through the dynamic attention mechanism, features at different levels are weighted and fused to enhance the sensitivity to key nodes and relationships and improve the ability to identify potential risks and opportunities.

[0156] (4) Real-time dynamic monitoring: With real-time update capabilities, it can quickly respond to dynamic changes in corporate relationships and provide a scientific basis for corporate decision-making.

[0157] The method of this application mainly evaluates the potential risks of enterprises in supply chain, technical cooperation, capital connection, market competition, legal compliance, market operation and systemic risks, that is, these risks may affect their business or damage their assets and reputation due to supply chain, cooperation, competition and other relationships; opportunities are identified as those that can bring positive impact and value to the enterprise, and identify and evaluate those situations and conditions that can bring potential growth, profit or strategic advantages to the enterprise, so as to stay ahead in the competitive market. This method breaks through the technical bottlenecks of traditional enterprise intelligence systems in dynamic modeling, multi-layer network representation and topological feature extraction. By combining the time-series-aware graph neural network with the theory of persistent homology in computational topology, accurate modeling of enterprise relationship networks in both time and structure is achieved. This innovative technical solution can not only capture the dynamic evolution pattern of enterprise relationships, but also reveal the potential risks contained in the network structure through topological feature analysis, providing a more systematic and in-depth analysis framework for enterprise risk management.

[0158] The following is an exemplary description of the enterprise risk assessment device based on dynamic graph neural network provided by this application.

[0159] like Figure 3 As shown, the embodiment of the present application provides an enterprise risk assessment device based on a dynamic graph neural network, and the enterprise risk assessment device 300 based on a dynamic graph neural network includes:

[0160] The acquisition module 301 is used to acquire the relationship data of multiple target enterprises; the relationship data of each target enterprise is used to describe the enterprise relationship between the target enterprise and other target enterprises, and the enterprise relationship is one of the multiple relationships;

[0161] A construction module 302 is used to construct a multi-layer enterprise social network based on all the relationship data; the multi-layer enterprise social network corresponds to the multiple relationships one by one, the multiple nodes in the multi-layer enterprise social network correspond to the multiple target enterprises one by one, the edge between two nodes in the same layer is the enterprise relationship between the corresponding two target enterprises, and the edge between two nodes in different layers is the influence relationship between the corresponding two target enterprises;

[0162] The feature extraction module 303 is used to extract features from the multi-layer enterprise social network using a dynamic graph neural network to obtain short-term feature embedding, long-term feature embedding and topological feature embedding of the multi-layer enterprise social network; the short-term feature embedding is used to describe the enterprise relationship information in the multi-layer enterprise social network over a short time span, the long-term feature embedding is used to describe the enterprise relationship information in the multi-layer enterprise social network over a long time span, and the topological feature embedding is used to describe the topological structure information of the multi-layer enterprise social network;

[0163] The fusion module 304 is used to obtain persistent features based on the multi-layer enterprise social network, and fuse the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features; the persistent features are used to describe the stability of each edge in the multi-layer enterprise social network;

[0164] The risk assessment module 305 is used to perform risk assessment on each target enterprise according to the fusion characteristics and the multi-layer enterprise social network to obtain a risk assessment result for each target enterprise; the risk assessment result is used to describe the risk level of the target enterprise.

[0165] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0166] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0167] like Figure 4 As shown, an embodiment of the present application provides a terminal device. The terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 4 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above-mentioned method embodiments when executing the computer program D102.

[0168] Specifically, when the processor D100 executes the computer program D102, it obtains the relationship data of multiple target enterprises, then constructs a multi-layer enterprise social network based on all the relationship data, and then uses a dynamic graph neural network to extract features from the multi-layer enterprise social network to obtain short-term feature embedding, long-term feature embedding and topological feature embedding of the multi-layer enterprise social network, and then obtains persistent features based on the multi-layer enterprise social network, and fuses the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features, and finally performs risk assessment on each of the target enterprises based on the fused features and the multi-layer enterprise social network to obtain the risk assessment results of each of the target enterprises. Among them, constructing a multi-layer enterprise social network can fully describe the association relationship between target enterprises, realize accurate modeling of enterprise relationships, fuse the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features, improve the information richness and comprehensiveness of the fused features, and perform risk assessment based on the information-rich fused features and accurate multi-layer enterprise social networks, which can effectively improve the accuracy of enterprise risk assessment.

[0169] The processor D100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0170] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card (Flash Card), etc. equipped on the terminal device D10. Further, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0171] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0172] An embodiment of the present application provides a computer program product. When the computer program product runs on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0173] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the enterprise risk assessment method device / terminal device based on dynamic graph neural network, recording medium, computer memory, read-only memory (ROM, Read-OnlyMemory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0174] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0175] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0176] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for enterprise risk assessment based on dynamic graph neural network, characterized in that: include: Obtain relationship data of multiple target companies; The relationship data of each target enterprise is used to describe the enterprise relationship between the target enterprise and other target enterprises, and the enterprise relationship belongs to one of multiple relationships; Construct a multi-layer enterprise social network based on all relationship data; the multi-layer enterprise social network corresponds one-to-one to multiple relationships, multiple nodes in the multi-layer enterprise social network correspond one-to-one to multiple target enterprises, the edge between two nodes in the same layer is the enterprise relationship between the corresponding two target enterprises, and the edge between two nodes in different layers is the influence relationship between the corresponding two target enterprises; Using a dynamic graph neural network to extract features from the multi-layer enterprise social network, short-term feature embedding, long-term feature embedding and topological feature embedding of the multi-layer enterprise social network are obtained; the short-term feature embedding is used to describe enterprise relationship information in the multi-layer enterprise social network over a short time span, the long-term feature embedding is used to describe enterprise relationship information in the multi-layer enterprise social network over a long time span, and the topological feature embedding is used to describe topological structure information of the multi-layer enterprise social network; Acquire persistent features based on the multi-layer enterprise social network, and fuse the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features; the persistent features are used to describe the stability of each edge in the multi-layer enterprise social network; Performing risk assessment on each of the target enterprises according to the fusion features and the multi-layer enterprise social network to obtain a risk assessment result for each of the target enterprises; The risk assessment results are used to describe the risk level of the target enterprise; The step of constructing a multi-layer enterprise social network based on all relationship data includes: For each relationship, target enterprises that have the relationship in the relationship data of all target enterprises are taken as hierarchical target enterprises of the relationship, and a single-layer enterprise social network of the relationship is generated according to the relationship data of all hierarchical target enterprises; Calculate the influence relationship between every two different single-layer enterprise social networks, and generate edges between nodes in different single-layer enterprise social networks based on all influence relationships to obtain a multi-layer enterprise social network; The step of fusing the persistent feature, the short-term feature embedding, the long-term feature embedding and the topological feature embedding to obtain the fused feature includes: Embedding the short-term feature and fusing the persistent feature to obtain a first feature; Embedding and fusing the first feature and the long-term feature to obtain a second feature; The second feature and the topological feature are embedded and integrated to obtain a fusion feature.

2. The enterprise risk assessment method according to claim 1, characterized in that: The embedding of the short-term feature and the fusion of the persistent feature to obtain the first feature includes: By formula: Calculate the first feature ; in, represents the dynamic attention weight, Indicates The first node feature of the node, represents the first The short-term characteristics of nodes, Represents a persistent feature, represents the scoring function, represents the first The short-term characteristics of nodes, , Representation Node The neighbor nodes of Represents a collection of nodes in a multi-layer enterprise social network.

3. The enterprise risk assessment method according to claim 2, characterized in that: The step of fusing the first feature with the long-term feature embedding to obtain a second feature includes: By formula: Calculate the second feature ; in, Indicates The second node feature of the node, Indicates the first The node and The weights between nodes, Indicates the first The neighboring nodes of a node, represents the enhanced long-term feature embedding, represents the first The components corresponding to the nodes, represents the first The components corresponding to the nodes; The embedding and integrating the second feature and the topological feature to obtain a fusion feature includes: By formula: Calculate fusion features ; in, represents a graph neural network, represents the weight, represents the enhanced topological features, Indicates the topological features The components corresponding to the nodes, Indicates the topological features The component corresponding to each node.

4. The enterprise risk assessment method according to claim 1, characterized in that: The step of performing risk assessment on each target enterprise according to the fusion feature and the multi-layer enterprise social network to obtain a risk assessment result for each target enterprise includes: For each target enterprise, perform the following steps: Calculate the centrality score of the node corresponding to the target enterprise according to the multi-layer enterprise social network; Based on the centrality score and the fusion feature, risk assessment is performed on the target enterprise to obtain a risk assessment result of the target enterprise.

5. The enterprise risk assessment method according to claim 1, characterized in that: After the step of acquiring persistent features based on the multi-layer enterprise social network and fusing the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features, the enterprise risk assessment method further includes: For each of the target enterprises, a comprehensive opportunity score of each of the target enterprises is calculated based on the multi-layer enterprise social network and the persistent features; the comprehensive opportunity score is used to describe the probability that the target enterprise has a potential development opportunity.

6. An enterprise risk assessment device based on dynamic graph neural network, characterized in that: include: An acquisition module, used to acquire relationship data of multiple target enterprises; The relationship data of each target enterprise is used to describe the enterprise relationship between the target enterprise and other target enterprises, and the enterprise relationship belongs to one of multiple relationships; A construction module is used to construct a multi-layer enterprise social network based on all relationship data; the multi-layer enterprise social network corresponds one-to-one to multiple relationships, multiple nodes in the multi-layer enterprise social network correspond one-to-one to multiple target enterprises, an edge between two nodes in the same layer is the enterprise relationship between the corresponding two target enterprises, and an edge between two nodes in different layers is the influence relationship between the corresponding two target enterprises; A feature extraction module, configured to extract features from the multi-layer enterprise social network using a dynamic graph neural network, and obtain short-term feature embedding, long-term feature embedding, and topological feature embedding of the multi-layer enterprise social network; the short-term feature embedding is used to describe enterprise relationship information in the multi-layer enterprise social network over a short time span, the long-term feature embedding is used to describe enterprise relationship information in the multi-layer enterprise social network over a long time span, and the topological feature embedding is used to describe topological structure information of the multi-layer enterprise social network; A fusion module, used to obtain persistent features based on the multi-layer enterprise social network, and fuse the persistent features, short-term feature embedding, long-term feature embedding and topological feature embedding to obtain fused features; the persistent features are used to describe the stability of each edge in the multi-layer enterprise social network; A risk assessment module, used to perform risk assessment on each of the target enterprises according to the fusion characteristics and the multi-layer enterprise social network, and obtain a risk assessment result for each of the target enterprises; the risk assessment result is used to describe the risk level of the target enterprise; The building blocks are specifically used to implement: For each relationship, target enterprises that have the relationship in the relationship data of all target enterprises are taken as hierarchical target enterprises of the relationship, and a single-layer enterprise social network of the relationship is generated according to the relationship data of all hierarchical target enterprises; Calculate the influence relationship between every two different single-layer enterprise social networks, and generate edges between nodes in different single-layer enterprise social networks based on all influence relationships to obtain a multi-layer enterprise social network; The fusion module is specifically used to implement: Embedding the short-term feature and fusing the persistent feature to obtain a first feature; Embedding and fusing the first feature and the long-term feature to obtain a second feature; The second feature and the topological feature are embedded and integrated to obtain a fusion feature.

7. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the enterprise risk assessment method based on dynamic graph neural network as described in any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the enterprise risk assessment method based on a dynamic graph neural network as described in any one of claims 1 to 5 is implemented.

Images