Industrial chain integrity evaluation model, system and method

Abstract

The invention discloses an industrial chain integrity assessment model, system and method, and relates to the technical field of industrial chain assessment, the model comprises a node feature coding module, a candidate node set selection module, a pairwise code generation module and an assessment module; the node feature coding module is used for learning feature representation of nodes in the industry chain recessive heterogeneous graph based on a message passing neural network; the candidate node set selection module is used for screening nodes which have obvious influence on the target node pair from the industry chain recessive heterogeneous graph based on the PPR score; the pairwise code generation module is used for calculating pairwise codes of a target node pair based on a graph attention mechanism and capturing a complex relationship between nodes; and the evaluation module is used for predicting the existence probability of the target node pair through the multi-layer perceptron and carrying out quantitative evaluation on the predicted existence probability through the integrity evaluation index. According to the method, the complex relation between the nodes in the industry chain recessive heterogeneous graph is accurately captured, and the influence relation between the nodes can be comprehensively quantified.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial chain evaluation, and in particular to an industrial chain integrity evaluation model, system and method based on a graph attention mechanism and a message passing network. Background Art

[0002] The integrity of the industrial chain covers the interconnection and coordination of each link in the industrial chain, as well as the overall resilience and recovery capacity. This concept not only involves the interdependence of each link within the industrial chain, but also includes the ability to respond to external shocks and interference.

[0003] Specifically, the resistance of an industrial chain refers to its ability to cope with various shocks and disturbances. When an industrial chain is hit by a shock, it can minimize the impact and maintain normal operations. In the modern economy, industrial chains are often faced with adverse effects from internal and external factors such as natural disasters, market fluctuations, and supply chain disruptions. Therefore, strong resistance is essential to reduce the losses caused by these risks and ensure the sustainable development of the industrial chain.

[0004] The resilience of the industrial chain emphasizes the ability of the industrial chain to quickly return to normal operation after suffering a shock, and even to turn crises into development opportunities. This ability is crucial for responding to emergencies and crises, as it can minimize production interruptions and losses and promote rapid adaptation and upgrading of the industrial chain. An industrial chain with good resilience can better adapt to changes and create more value and competitive advantages.

[0005] The integrity of the industrial chain is not only an important factor in ensuring the stable operation of the industrial chain, but also a key element in achieving coordinated regional economic development and improving national competitiveness. By improving the resistance and recovery capabilities of the industrial chain, the stability of the economic system can be better maintained and improved, while providing strong support for sustained growth and innovation. Therefore, the assessment and improvement of the integrity of the industrial chain has become an indispensable part of modern economic management and strategic planning. With the increasing complexity and interconnectedness of the global supply chain, enterprises are not only affected by internal factors, but also by external factors such as fluctuations in the global market, natural disasters, and political events. For this reason, the assessment of the integrity of the industrial chain has become an indispensable link, aiming to ensure the smooth operation of the entire industrial chain, enhance the resilience of the industrial chain, reduce potential risks, and improve the response capabilities of enterprises.

[0006] In current research, there is a relative lack of literature related to the assessment of industrial chain integrity, and there is almost no quantitative research on industrial chain integrity. Existing literature often uses a relatively macro perspective to analyze the industrial chain, but rarely provides specific, quantifiable descriptions to assess the level of industrial chain integrity. To some extent, the formation process of the industrial chain is actually the formation process of the supply chain, accompanied by the formation of the value chain. Its core lies in the supply and demand relationship between different enterprises. Therefore, if we want to conduct an in-depth study of the integrity of the industrial chain, we inevitably need to pay attention to an important part of the industrial chain: the supply chain.

[0007] The supply chain is a core component of the industrial chain, which undertakes tasks such as managing logistics, resource flow, and the delivery of products or services. In practice, the management and optimization of the supply chain is of great significance to ensure the integrity and resilience of the industrial chain. By establishing a transparent supply chain network, optimizing logistics and inventory management, establishing supplier relationships, and implementing risk management strategies, enterprises can improve the efficiency and reliability of their supply chains, thereby enhancing the overall integrity of the industrial chain. However, in actual assessments, there are problems such as insufficient mutual trust between enterprises, fierce competitive relations, and protection of commercial secrets, which make it difficult to obtain key information in the supply chain, thus posing a challenge to the integrity assessment of the entire industrial chain.

[0008] Previous studies have mainly used traditional methods such as the Delphi method, system analysis method, and interview method to improve the visibility of supply chain information. In addition, some technical means such as companies sending questionnaires, using intermediaries to perform supply chain monitoring, and deploying tracking technologies (including RFID, etc.) can also help improve the visibility of hidden information. However, these methods are often affected by the experience and bias of practitioners and researchers, which may lead to insufficient objectivity and poor repeatability of the evaluation. In addition, these methods usually require companies to cooperate in sharing information, and the accuracy and completeness of the information cannot be guaranteed. Due to the lack of mutual trust, suppliers may intentionally not disclose strategically important relationships, operations, and environmental practices. Summary of the invention

[0009] The present invention aims to provide an industrial chain integrity assessment model, system and method based on graph attention mechanism and message passing network, which can solve the above problems.

[0010] In order to solve the above problems, the technical solution adopted by the present invention is as follows:

[0011] In a first aspect, the present invention provides an industrial chain integrity assessment model, including a node feature encoding module, a candidate node set selection module, a pair coding generation module and an assessment module;

[0012] The node feature encoding module is used to learn the feature representation of nodes in the implicit heterogeneous graph of the industrial chain based on the message passing neural network to obtain the feature encoding of the nodes;

[0013] The candidate node set selection module is used to select nodes that have a significant impact on the target node from the implicit heterogeneous graph of the industrial chain according to the feature coding of the node and based on the PPR score to obtain a candidate node set;

[0014] The paired coding generation module is used to combine the feature coding of the candidate nodes and the relative position coding of the candidate nodes relative to the target node pair, calculate the paired coding of the target node pair based on the graph attention mechanism, and capture the complex relationship between the nodes;

[0015] The evaluation module is used to predict the existence probability of the target node pair through a multi-layer perceptron, a comprehensive candidate node set, feature coding of the target node, and paired coding of the target node pair, and to quantitatively evaluate the predicted existence probability through an integrity evaluation index.

[0016] As a further description of the above technical solution, the node feature encoding module uses a graph convolutional network as an encoder to perform feature encoding on each node in the implicit heterogeneous graph of the industrial chain; the graph convolutional network uses the ReLU function as an activation function; the graph convolutional network contains multiple graph convolutional layers, each layer updates the feature representation of the node and can capture neighborhood information of different scales; the output of each layer becomes the input of the next layer, allowing information to propagate throughout the graph structure, and after learning through multi-layer message passing, node i learns the node feature encoding containing neighbor information and global information.

[0017] As a further description of the above technical solution, before encoding the implicit heterogeneous graph of the industrial chain, it needs to be preprocessed, including: representing the implicit heterogeneous graph of the industrial chain as an adjacency matrix and a node feature matrix; adding self-loops to each node in the adjacency matrix; and normalizing the adjacency matrix.

[0018] As a further description of the above technical solution, the PPR score calculation formula used by the candidate node set selection module is:

[0019]

[0020] Among them, PPR(i,k) represents the personalized PageRank value from node i to node k, α is the damping factor, represents the set of neighbor nodes of node j, deg(k) represents the out-degree of node k, δ ij is an indicator function. When i=j, δ ij =1, otherwise δ ij =0.

[0021] As a further description of the above technical solution, assuming that there is a target node pair (a, b) in the implicit heterogeneous graph of the industrial chain, the method of screening out the nodes that have a significant impact on it includes:

[0022] Calculate the personalized PageRank value ppr(a,u) from node a to node u, and the personalized PageRank value ppr(b,u) from node b to node u, where node u is the root node in the implicit heterogeneous graph of the industrial chain;

[0023] For a certain type of neighbor node of the target node pair (a, b) in the implicit heterogeneous graph of the industrial chain, a PPR threshold η is set for it, and then all nodes whose PPR scores relative to the two nodes a and b in the target node pair (a, b) are retained, that is, the nodes with significant influence on the target node pair (a, b) are obtained;

[0024] The specific selection implementation formula for filtering node sets of various nodes that have significant impact on the target node pair (a, b) is as follows:

[0025]

[0026] in, is the filtered node set of neighbor nodes of node type π, η π is the PPR threshold corresponding to the neighbor node type π.

[0027] As a further description of the above technical solution, the paired encoding formula used by the paired encoding generation module is:

[0028]

[0029] Among them, w(a,b,u) measures the importance of node u to the target node pair (a,b), h(a,b,u) is the feature encoding of node u relative to (a,b), V is the set of all nodes in the graph, and ⊙ is the Hadamard product, which means the multiplication of corresponding elements of two vectors.

[0030] As a further description of the above technical solution, the feature encoding h(a,b,u) of node u relative to (a,b) is defined as:

[0031]

[0032] Among them, h u Table node u feature encoding, represents the relative position encoding of node u relative to the target node pair (a, b), ∥ is the connection operator, and W is a learnable weight matrix used to perform linear transformation on the concatenated vector.

[0033] The importance w(a,b,u) of node u to the target node pair (a,b) is modeled by attention as follows:

[0034]

[0035] Among them, h a ,h b Represent the feature codes of nodes a and b respectively; represents the set of nodes remaining after removing nodes a and b in the graph; φ(·) is the GATv2 attention mechanism; is the unnormalized attention weight of node u for (a, b); w(a, b, u) is the normalized attention weight obtained after using the softmax function.

[0036] As a further description of the above technical solution, the relative position encoding calculation formula of node u relative to the target node pair (a, b) is:

[0037] rpe (a,b,u) =MLP(ppr(a,u),ppr(b,u))

[0038]

[0039] Among them, MLP is a multi-layer perceptron, which is used to parameterize the relative position encoding calculation method.

[0040] As a further description of the above technical solution, the prediction score function used by the evaluation module is:

[0041]

[0042] Where p(a,b) is the probability of predicting that there is a link between node a and node b; σ is the sigmoid function; ⊙ is the Hadamard product; ∥ is the join operator; h a represents the feature code of node a; s(a,b) represents the pairwise coding information formed by the node pair (a,b); and Belonging to the candidate node set, they represent the number of common neighbors, the number of one-hop neighbors, and the number of neighbors greater than one hop of the node pair (a, b).

[0043] As a further description of the above technical solution, the calculation formula of the industry chain integrity score is as follows:

[0044]

[0045] Among them, RT(C) represents the resistance measurement of the industrial chain, and RL(C) represents the recovery measurement of the industrial chain.

[0046] In a second aspect, the present invention provides an industrial chain integrity assessment system, which is embedded in the industrial chain integrity assessment model described in the first aspect.

[0047] In a third aspect, the present invention provides an industrial chain integrity assessment method, which assesses the integrity of an industrial chain based on the industrial chain integrity assessment model described in the first aspect.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1) The present invention introduces message passing neural network and graph attention mechanism to accurately capture the complex relationship between nodes in the implicit heterogeneous graph of the industrial chain, effectively solving the problems of poor information visibility, strong subjectivity and poor repeatability in traditional methods. By screening key nodes through PPR scoring and generating paired codes, the influence relationship between nodes can be fully quantified, which improves the objectivity and accuracy of the evaluation. In addition, the multi-layer perceptron combines multi-dimensional features to quantitatively evaluate the integrity of the industrial chain, overcoming the limitation of existing research that lacks specific and quantifiable descriptions, and providing a scientific basis for the optimization and risk management of the industrial chain.

[0050] 2) The implicit heterogeneous graph of the industrial chain is used to abstractly represent the complex structure of the industrial chain, which can adapt to the incomplete information and implicit connections that are common in real industrial chain data; the implicit heterogeneous graph of the industrial chain regards the enterprises, products and interactions in the industrial chain as nodes and edges in a heterogeneous graph, so that the model can intelligently infer those unclear or missing connections, which not only enhances our understanding of the complex dynamics within the industrial chain, but also improves our ability to identify and fill information gaps, thereby more accurately revealing the key connections and potential synergies in the industrial chain.

[0051] 3) The industry chain resistance capacity index and industry chain resilience index related to integrity on the industry chain graph dataset; these indicators provide standards for the evaluation of the industry chain, which helps to understand the operation of the industry chain and evaluate its resistance and recovery capacity. To a certain extent, it fills the research gap in this field and provides a scientific method for evaluating the integrity of the industry chain.

[0052] 4) The link prediction method based on the graph attention mechanism and the message passing neural network in the model of the present invention characterizes the relevant indicators of the integrity of the industrial chain: the link prediction based on the attention mechanism and the message passing neural network is very suitable for industrial chain analysis because it can comprehensively consider various structural information in the industrial chain, including local structural information, global structural information, and feature similarity. In this way, the model of the present invention can learn rich feature representations between enterprises and products, and better capture the correlation between them. In the context of the industrial chain, this means that the model can identify and predict potential connections between enterprises and between enterprises and products, which is crucial for understanding the complex dynamics of the industrial chain and optimizing the structure of the industrial chain. Especially in the face of incomplete information in the industrial chain, this link prediction method can improve the understanding of key connections in the industrial chain by intelligently filling in information gaps. Therefore, the model of the present invention not only improves the accuracy of link prediction, but also provides strong support for industrial chain management. The performance of the model in the real industrial chain data set surpasses other advanced graph neural network methods, which shows the effectiveness and practicality of the model, and it provides a higher level of performance for the evaluation of industrial chain integrity.

[0053] 5) Existing pairwise coding methods have limitations, often only modeling some potential factors such as local structure, ignoring other key factors such as node attributes, global structure and feature similarity between nodes. In addition, existing methods often adopt a unified pairwise coding strategy, using the same combination of prediction factors for all links, lacking adaptability to different link characteristics. In the industrial chain dataset, the links between different industries and products may require different combinations of prediction factors. The method of the present invention that can adaptively model multiple link prediction factors is crucial to improving the prediction accuracy of the model in the industrial chain dataset.

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the embodiments of the present invention are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0056] Figure 1 It is the overall process of the industrial chain integrity assessment model described in the embodiment;

[0057] Figure 2 are the heterogeneous graph and the implicit heterogeneous graph described in the embodiments;

[0058] Figure 3 Flow chart of the candidate node set selection module in the embodiment;

[0059] Figure 4 Flow chart of the paired coding generation module in the embodiment;

[0060] Figure 5 Flowchart of the evaluation module in an embodiment. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0062] The embodiment of the present invention provides an industrial chain integrity assessment model based on a graph attention mechanism and a message passing network, referring to Figure 1 As shown in the figure, it includes four parts: node feature encoding module, candidate node set selection module, pair coding generation module and evaluation module. The industrial chain integrity evaluation model can predict the links in the implicit heterogeneous industrial chain network diagram, and quantitatively evaluate the prediction results through integrity evaluation indicators to achieve integrity evaluation on the industrial chain data set. The specific instructions are as follows:

[0063] 1. Parameter definition

[0064] Node pair: In a graph G = (V, E), any two nodes u and v form an ordered pair (u, v), where V is the set of all nodes in the graph and E is the set of all edges in the graph. If (u, v) ∈ E, then u and v are called a directly connected node pair; if there is no direct edge between u and v, but they can reach each other through other nodes and edges in the graph, then they are called an indirectly connected node pair.

[0065] Paired encoding: Paired encoding is a technique for capturing the relationship between node pairs. It encodes node pairs to generate feature representations that reflect the relationship between node pairs. This method is very useful when processing graph-structured data, especially in tasks that require considering complex relationships between node pairs, such as knowledge graph completion and molecular pair similarity calculation. The core idea of ​​pair encoding is to encode each pair of nodes (v i ,v j ) is mapped to a feature vector, which represents the relationship between the pair of nodes. i ,v j ), pairwise encoding generates a feature vector e ij , the vector can contain the node v i and vj The characteristics of the edges between them are expressed as e ij =f(h i ,h j ,a ij ), where h i and h j They are node v i and v j The characteristic representation of a ij is the feature of the edge between two nodes, and f is an encoding function. The encoding function can be a simple connection operation or a more complex neural network model. For example, a multi-layer perceptron (MLP) can be used to encode node pairs: ij =MLP([h i ‖h j ‖a ij ]), where (||) represents the concatenation operation of vectors.

[0066] Heterogeneous graph: Figure 2 As shown in (a), a heterogeneous graph is a data structure consisting of a node set V and a link set E, expressed as G = (V, E). The heterogeneous graph is also associated with a node type mapping function φ: V → A and a link type mapping function Ψ: E → R. The node type mapping function φ maps each node in the graph to a type in a predefined node type set A, and the link type mapping function Ψ maps each edge in the graph to a type in a predefined edge type set R. In addition, |A| + |R| > 2.

[0067] Implicit heterogeneous graph: For industrial chain data, a large number of enterprise types and relationships are missing, that is, the types of nodes and link sets in the heterogeneous graph are missing. This invention defines such a heterogeneous graph as an implicit heterogeneous graph, see Figure 2 (b) shows that, for G = (V, E), there are missing node type sets A and link type sets R, as well as missing mapping relations φ and Ψ. An extreme case is that the known node type set A and link type set R are both empty sets, in which case the observed G′ = (V, E) is actually an isomorphic graph.

[0068] PPR: The PPR score is a random walk-based algorithm that measures the relative importance of nodes in a graph and predicts whether there is a potential link between two nodes in the graph. It is a variant of the PageRank algorithm, which was originally used by the founders of Google to rank web pages to determine the importance or quality of web pages. The PPR score takes into account user preferences, allowing users to start from designated seed nodes and calculate the relative importance of these seed nodes and other nodes in the graph. The PPR score takes into account the transition probability from the designated seed node to other nodes, as well as a parameter α called the "jump probability". This parameter allows a certain probability of jumping out of the current path and visiting any node in the graph during the random walk. This design allows the PPR score to capture local relationships while also taking into account the global structure of the graph. The goal of the PPR score is to calculate the relative importance of all nodes relative to the root node u. Start walking from the node corresponding to the root node u, stop walking at each node with a probability of 1-α and start again from the root node u, or continue walking with a probability of α and randomly select a node from the nodes pointed to by the current node according to a uniform distribution to walk downstream. In this way, after many rounds of walking, the probability of each vertex being visited will converge and become stable, and then the probability can be used for ranking.

[0069] The PPR score of the root node i and the target node j with the transition probability α is expressed as ppr(i,j), which indicates the relative importance of node j to the root node i when starting from the root node i. The calculation of the PPR score is based on the following formula:

[0070]

[0071] Among them, PPR(i,j) represents the personalized PageRank value from node i to node j. α is a damping factor, usually between 0 and 1, which represents the probability of continuing to walk along the edge of the network during a random walk, and a common value is 0.85. represents the set of neighbor nodes of node j. deg(k) represents the out-degree of node k. δ ij is an indicator function. When i=j, δ ij =1, otherwise δ ij =0.

[0072] Relative Position Encoding (RPE): Relative position encoding is a position encoding method in the Transformer model. It is different from traditional absolute position encoding. RPE focuses on the relative distance between elements in the sequence rather than their absolute position in the sequence. When exploring the application of relative position encoding in industrial chain datasets, the goal of this invention is to capture the positional relationship between nodes so as to differentially encode different types of information (local information and global information). RPE can provide a way for the model to understand the relative positional relationship between enterprises and products in the industrial chain, which is crucial to revealing the complex structure of the industrial chain. However, there are challenges in implementing RPE. An effective method is to use the dual radius node labeling (DRNL) technology, but this method has a high computational cost on large graph datasets. In addition, traditional RPE methods face difficulties in processing large-scale graphs because they usually rely on pairwise distances or eigenvectors of the Laplacian operator, and these calculations become impractical on large graphs.

[0073] The present invention designs a RPE based on the PPR score, which can not only distinguish the relationship between different nodes and the target link, but also can be calculated efficiently. Because the goal of the PPR score is to start from the root node u and calculate the relative importance of all nodes in the graph relative to the root node u. And PPR sets a transition probability α so that there is a certain probability of jumping out of the current path and accessing any node in the graph during the random walk. This design enables the PPR score to capture local relationships and moderately consider the global structure of the graph. Therefore, the relative importance between nodes calculated by the PPR score can be used as the relative position encoding between node pairs. Because the PPR score can measure the relative importance between nodes, the PPR score can be used as an intuitive and useful method to understand the structural relationship between node u and the two nodes in the target node pair (a, b). If both ppr(a,u) and ppr(b,u) scores are very high, it means that starting from node a or node b, it only needs to be connected to node u through a shorter random walk path. This means that node u has a strong influence on the nodes in the target node pair (a, b), that is, it has a strong relative importance. If both PPR scores are low, it indicates that the relationship between node u and the target node pair (a, b) is likely small. Therefore, the PPR score provides a convenient way to distinguish how nodes are structurally related to the target node pair.

[0074] Because the goal of the PPR score is to calculate the relative importance of all nodes in the graph relative to the root node u, starting from the root node u. In order to calculate the RPE of node u relative to the target node pair (a, b), the present invention uses the PPR scores of node u relative to the two nodes (a, b) in the target node pair (a, b), that is, ppr(a,u) and ppr(b,u). In order to better capture the complex relationship between nodes, the present invention uses MLP to parameterize the relative position encoding calculation method, that is, by feeding ppr(a,u) and ppr(b,u) as input into MLP to obtain the relative position encoding value of node u relative to the target node pair (a, b). The calculation formula can be expressed as

[0075] rpe (a,b,u) =MLP(ppr(a,u),ppr(b,u)) (2)

[0076] In many graph structured data, the order of nodes does not affect the relationship between them. In order to ensure that the relative position encoding does not change the order of nodes in the target node pair, that is, the relative position encoding of (a, b, u) and (b, a, u) is the same, the present invention further sets RPE to be equal to the sum of the representations given by (a, b, u) and (b, a, u), avoiding the model's sensitivity to the node order, thereby improving the generalization ability of the model. The calculation formula is as follows:

[0077]

[0078] is to use formula (2) to calculate rpe (a,b,u) and rpe (b,a,u) The result obtained by post-summing ensures the order invariance of the nodes in the target node pair by summing the relative position coding. Considering that formula (3) may not be able to completely distinguish the three types of relative position coding of the target node pair (a, b) CN (common neighbors), a and b's 1-hop neighbors, and a and b's >1-hop neighbors, the present invention uses three independent MLPs, that is, in formula (2), for the target node pair (a, b) CN, a and b's 1-hop neighbors, and a and b's >1-hop neighbors, different MLPs are used to parameterize the calculation formula. By using different MLPs for each case, the model can learn how to distinguish these different types of nodes.

[0079] Industry chain resistance measurement: In the industry chain, the industry chain resistance measurement is specifically the number of connections between enterprises and other enterprises. The stronger the resistance, the more connections the enterprise has. In the industry chain C = (Z, R) there is a set of enterprises Z = {z1, z2, ..., z n}, the inter-enterprise link relationship set R = {r1, r2, …, r m}, let A(C)=[zi,j ] n×n , The resistance of the industrial chain is defined as:

[0080]

[0081] Industrial chain resilience measurement: For an industrial chain C = (Z, R), there is a set of hidden relationships between enterprises in the industrial chain R′ = {r m+1 ,r m+2 ,…,r m+k}, the more hidden links a chain has, the stronger its resilience is. The resilience metric of the industry chain is defined as:

[0082]

[0083] 2. Node feature encoding module

[0084] The node feature encoding module is used to learn the feature representation of nodes in the implicit heterogeneous graph of the industrial chain based on the message passing neural network, and obtain the feature encoding of the nodes, as follows:

[0085] In the implicit heterogeneous network of the industrial chain, there are different kinds of relationships between nodes, and these relationships may carry different meanings. In order to better understand these relationships, this embodiment first uses MPNN (i.e., message passing neural network, which is not a specific model, but a general framework containing the commonalities of a series of graph neural network models) to extract features from the nodes in the implicit heterogeneous graph of the industrial chain, generates an encoding for each node, and learns the exchange of information between the surrounding enterprise nodes through the message passing neural network, so that the feature encoding of the node can learn the feature information of the surrounding nodes.

[0086] In the node feature encoding module, this embodiment uses a graph convolutional network (GCN) as an encoder to perform feature encoding on each node in the industrial chain.

[0087] First, the implicit heterogeneous graph of the industrial chain needs to be represented as the adjacency matrix A and the node feature matrix X.

[0088] The adjacency matrix A is an N×N matrix, where N is the number of nodes and A ij Indicates whether there is an edge between node i and node j. The node feature matrix X is an N×D matrix, where D is the feature dimension of each node.

[0089] In order to allow each node to take its own characteristics into account in the convolution operation, it is necessary to add self-loops to each node in the adjacency matrix. This can be achieved by adding the identity matrix I to the adjacency matrix A, that is,

[0090] Then the adjacency matrix is ​​normalized. Normalized adjacency matrix The calculation formula is: in yes The degree matrix of

[0091] After completing the preprocessing of the graph dataset (i.e., the implicit heterogeneous graph of the industrial chain), the GCN model is constructed. The core operation of GCN is graph convolution, and its basic form can be expressed as:

[0092]

[0093] Among them, H l represents the node feature matrix of the lth layer, H (l+1) represents the output feature matrix of layer l+1, W l is the weight matrix of the lth layer, σ is the nonlinear activation function, and the ReLU function is used as the activation function here. The GCN model can contain multiple graph convolutional layers, each of which updates the feature representation of the node and can capture neighborhood information of different scales. The output of each layer becomes the input of the next layer, allowing information to propagate throughout the graph structure. After learning through multi-layer message passing, node i will learn the node feature encoding h containing neighbor information and global information. i .

[0094] 3. Candidate node set selection module

[0095] Link prediction tasks face significant challenges when processing large-scale graph data, especially in complex datasets such as industrial chains. Directly predicting links for all possible node pairs is not only computationally expensive, but also dramatically increases the demand for memory. To address this problem, the present invention adopts a selective and sparse attention mechanism, which enables the model to focus on only a small number of nodes in the graph, thereby improving the efficiency of completing missing edges in the industrial chain.

[0096] In the candidate node set selection module, it is necessary to determine the target node pair (a, b) of interest and select the node set that is valuable to the target node pair (a, b). By selecting this small set of nodes, they can provide important contextual information for the target node pair (a, b) in a small number to best learn paired information. The goal of the PPR score is to calculate the relative importance of all nodes in the graph relative to the root node u starting from the root node u. Therefore, the PPR score can be used as a good tool to measure the impact of one node on another node. The present invention uses the PPR score to measure the relative importance of node u to the two nodes in the target node pair (a, b), where the calculation formula of the PPR score is shown in formula (1).

[0097] That is, for the target node pair (a, b), the CN of the target node pair (a, b), the 1-hop neighbors of a and b, and the PPR scores of node u in the >1-hop neighbors of a and b for nodes a and b are calculated respectively, that is, ppr(a,u) and ppr(b,u) are calculated, see Figure 3 As shown. The present invention sets a threshold for each type of neighbor node, and then retains all nodes whose PPR scores relative to the two nodes in the target node pair are higher than a certain threshold η, so as to select a set of nodes that are valuable to the target node pair (a, b). In specific implementation, the present invention selects different types of high-value nodes for the target node pair (a, b) by applying different thresholds to CN, 1-Hop and >1-Hop nodes. The specific selection implementation formula for the filtered node set of each type of node is as follows:

[0098]

[0099] in, is the filtered node set of neighbor nodes of node type π, η π is the PPR threshold corresponding to the neighbor node type π, where π∈{CN,1-Hop,>1-Hop}.

[0100] 4. Paired encoding generation module

[0101] In industrial chain datasets, link prediction tasks face significant challenges because these datasets contain numerous entities and complex relationship networks. To cope with this complexity, link prediction models need to be able to adapt to the diverse link types formed by multiple factors.

[0102] By introducing the attention mechanism, the present invention proposes a pairwise coding generation module that can widely cover potential link prediction factors and adaptively generate pairwise coding information for each target node pair. Figure 4 As shown, the paired encoding generation module comprehensively considers the three link prediction factors of local structure information, global structure information and feature similarity, and can achieve sufficient feature extraction for the target node pair (a, b).

[0103] The present invention proposes the following pair encoding formula:

[0104] s(a,b)=∑ u∈V w(a,b,u)⊙h(a,b,u) (8)

[0105] Among them, w(a,b,u) measures the importance of node u to the target node pair (a,b), h(a,b,u) is the feature encoding of node u relative to (a,b), V is the set of all nodes in the graph, and ⊙ is the Hadamard product, which means the multiplication of corresponding elements of two vectors.

[0106] We hope that for the target node pair (a, b), node u can learn the weight for the target node pair (a, b) based on the comprehensive consideration of local structural information, global structural information, feature similarity and other link prediction factors. The present invention uses softmax attention, which can dynamically learn the correlation between different nodes and the target node pair. In this way, for multiple target links, the contribution of different nodes can be emphasized, thereby flexibly modeling different LP factors. The attention here is between different sequences (i.e., target node pairs and nodes), so we can regard it as a form of cross attention.

[0107] In order to enhance the adaptability of pairwise encoding to various links, various types of information need to be merged and fused together. This enables the attention mechanism to identify and prioritize relevant information for each target link, facilitating effective modeling of various LP factors.

[0108] The present invention considers two types of information: node feature information and relative position information.

[0109] The node feature information includes the feature representations of the two nodes in the target node pair and the node being processed. The node features play a role in link formation and contain structural information features.

[0110] The relative position information reflects the relative position of node u and the target node pair (a, b) in the graph in the local and global structural context.

[0111] By considering node feature information and relative position information, the space of potential LP factors can be covered, that is, the three link prediction factors related to the existence of links, namely local structure information, global structure information and feature similarity, are comprehensively considered.

[0112] In this embodiment, the feature of node u is represented by h u , the relative position encoding (RPE) is expressed as

[0113] The node importance w(a,b,u) is modeled by attention as follows:

[0114]

[0115] in, represents the set of nodes remaining after removing nodes a and b in the graph; φ(·) is the GATv2 attention mechanism; is the unnormalized attention weight of node u for (a, b); w(a, b, u) is the normalized attention weight obtained after using the softmax function. The attention weight w(a, b, u) can be considered as the influence of node u on (a, b) relative to all nodes in G. This enables the model to emphasize different LP factors for each target node pair.

[0116] The node encoding h(a,b,u) contains the combination of the features of node u and RPE, defined as:

[0117]

[0118] where h u Table node u's characteristic expression, rpe (a,b,u) represents the position encoding of node u relative to (a, b), ∥ is the connection operator, and W is a learnable weight matrix used to perform linear transformation on the concatenated vector.

[0119] By calculating the attention weights w(a,b,u) and node features h(a,b,u) respectively, the paired encoding information s(a,b) of the node pair (a,b) can be calculated. This paired encoding s(a,b) not only integrates the local structural features in the industrial chain data set, but also integrates the global structural information, so that it can more comprehensively capture the complex link relationships between enterprises, between products, and between products and enterprises. The advantage of this encoding method is that it can adapt to the diversity and dynamics of industrial chain data, thereby effectively and accurately completing the missing edges in the industrial chain. In this way, the model can not only identify key nodes and potential connections in the industrial chain, but also predict new or missing links, providing data support for the optimization and expansion of the industrial chain.

[0120] 5. Evaluation Module

[0121] exist Figure 5 In the evaluation module shown, the present invention integrates the candidate node set, the feature coding of the target node, and the paired coding of the target node pair to predict the missing edges of the data set in the industrial chain and complete the completion of these missing edges.

[0122] In order to fully capture local and global structural information, this paper adopts a comprehensive method that combines node feature encoding, pairwise encoding, and structural information of node pairs, including the number of common neighbors, the number of 1-hop neighbors, and the number of neighbors greater than 1 hop. This method can not only reveal the direct connections between nodes, but also provide insights into more complex interactions and dependencies in the industrial chain. The scoring function is:

[0123]

[0124] Where p(a,b) is the probability of predicting that there is a link between node a and node b; σ is the sigmoid function, which is used to convert the output of MLP into a probability value; ⊙ is the Hadamard product, also known as the element-wise product, which means the multiplication of the corresponding elements of two vectors; ∥ is the concatenation operator, which is used to concatenate two vectors or matrices to form a new vector or matrix; h a represents the feature encoding of node a, s(a,b) represents the paired encoding information formed by the node pair (a,b), and They represent the number of common neighbors, the number of one-hop neighbors, and the number of neighbors greater than one hop for the node pair (a, b).

[0125] After obtaining the model prediction result p(a,b), a threshold is set by default. If the predicted probability value p(a,b) is greater than the threshold, it is considered that the target node pair (a,b) is connected by an edge, that is, the value is set to 1 in the adjacency matrix. After traversing all target links (i.e. target node pairs), the missing edge completion task of the industrial chain is completed.

[0126] Then, the resistance capacity RT(C) and recovery capacity RL(C) of the industrial chain are calculated according to the adjacency matrix;

[0127] Then, the integrity score of the industrial chain is calculated according to the industrial chain integrity score calculation formula. The higher the score, the stronger the integrity. The industrial chain integrity score calculation formula is as follows:

[0128]

[0129] Among them, RT(C) represents the resistance measurement of the industrial chain, and RL(C) represents the recovery measurement of the industrial chain.

[0130] 6. Experiment and analysis

[0131] This paper compares the industrial chain integrity evaluation model (LPAPE: LinkPredictor with Attention and Position Encoding) proposed in this paper with other state-of-the-art methods on real-world industrial chain datasets, and provides experimental settings and experimental results analysis. The prediction performance of the LPAPE model is evaluated on the Cora, Citeseer and Pubmed datasets, and the generalization performance of the model is further evaluated.

[0132] 6.1 Dataset

[0133] Table 1 Data scale of integrated circuit industry chain

[0134]

[0135] The integrated circuit industry chain data set used in this experiment covers the real financial data of 1,732 listed companies in China from 2018 to 2023 and 430 upstream and downstream product data compiled, totaling more than 40,000 data. Among them, the financial data includes 18 financial-related indicators, such as current ratio, return on net assets, debt-to-asset ratio, inventory turnover rate, etc., which reflect the financial and operating conditions of these listed companies. Through these data, the supply relationship and value link of the industrial chain can be deeply analyzed to discover the important features hidden in it. The data collection and collation work adopts a variety of methods. Among them, the financial data mainly comes from the RESSET financial research database, which is obtained through manual query and website download. Some financial indicator data are also supplemented and improved by the Wind database and Tianyancha to ensure the comprehensiveness and accuracy of the data.

[0136] In addition, for each listed company, the data set also includes a timestamp data, a corporate risk score data, and two corporate investment relationship data. The corporate risk score data mainly comes from the Wind database, which conducts a detailed assessment of the corporate operating risk at the current timestamp. These risk scores can reflect the company's competitive position in the market, financial stability, and future development potential. The two corporate investment relationship data mainly involve the number of investors and the number of outbound investments of the company. These data provide an important reference for in-depth mining of hidden competition or partnership information. The number of investors can reflect the company's financing capabilities and investor confidence, while the number of outbound investments can show the company's strategic expansion and industrial layout.

[0137] In addition, the upstream and downstream product data is an integrated circuit industry chain graph structure formed by combining upstream and downstream connection relationships. It intuitively reflects the complete appearance and internal mechanism of the industrial chain and is the main framework of the integrated circuit industry chain data set.

[0138] Table 2 Introduction to the scientific literature graph dataset

[0139]

[0140] As shown in Table 2, the Cora dataset contains 2708 scientific publications divided into seven categories, such as case-based, genetic algorithm, neural network, etc. Each publication is described by a 1433-dimensional 0 / 1 value word vector, indicating the absence / presence of the corresponding word in the dictionary. The citation network consists of 5429 edges, representing the citation relationship between papers. The Citeseer dataset consists of 3312 scientific publications divided into six categories, including Agents, AI, DB, IR, ML, and HCI. Each node has a 3703-dimensional binary feature vector, representing the content of the paper. The citation network contains 4732 edges, recording the citation or citation information between papers. The Pubmed dataset contains 19717 scientific publications on diabetes, divided into three categories, such as Diabetes Mellitus, Experimental, Diabetes Mellitus Type 1, and Diabetes Mellitus Type 2. Each publication is described by a TF / IDF weighted word vector in a dictionary consisting of 500 unique words. The citation network consists of 44,338 links, showing the citation relationships between papers.

[0141] Cora, Citeseer, and Pubmed are all well-known scientific literature graph datasets, containing research papers in the fields of computer science, computer science, and biomedicine, respectively. The nodes in these datasets represent papers, and the edges represent the citation relationships between papers. Node features are usually based on the text content of the papers and are used for graph machine learning tasks such as node classification and link prediction.

[0142] 6.2 Experimental Setup

[0143] Table 3 Experimental environment configuration

[0144]

[0145] In the experimental stage, we evaluated the performance of the model on a real industry chain dataset, using commonly used evaluation indicators for link prediction tasks, including Accuracy, Auprc, Auroc, and F1 score. Each method was adjusted to the optimal parameter settings on our dataset, and each training was more than 200 rounds. At the same time, each method was trained 5 times independently, and the average of the evaluation indicators was used as the final evaluation result of the model. The learning rate of our LinkPredictor with Attention and Position Encoding (LPAPE) model was set to 0.005, and the η in the node selection module was set to 0.01. π,π∈{CN,1-Hop,>1-Hop}, are set to 0.01, 0.01, 0.0001 respectively, the weight decay coefficient is 0.0001, and the data set is divided into training set, validation set, and test set according to 85:5:10 to obtain the best experimental results. All training and validation processes of the above experiments are completed through a server cluster and a personal computer. The experimental environment configuration is shown in 2-3.

[0146] Afterwards, in order to further evaluate the generalization performance and adaptability of the LPAPE model we proposed on various datasets, we conducted link prediction evaluation experiments on three scientific literature graph datasets, using the MRR (Mean Reciprocal Rank) indicator widely used in link prediction to evaluate the performance of the LPAPE model we proposed. The purpose of this step is to verify that the LPAPE model still has excellent prediction performance in the face of various graph data, and has excellent generalization performance and robustness.

[0147] 6.3 Comparison Method

[0148] For a comprehensive comparison, we compare it with 6 other state-of-the-art link prediction neural network models:

[0149] 1. GraphSAGE (Graph Sample and Aggregated) is a graph neural network method for node classification and link prediction. It learns the representation of nodes by sampling and aggregating the features of neighboring nodes.

[0150] 2. GIN (Graph Isomorphism Network) is a graph neural network with graph isomorphism properties that can maintain the invariance of node arrangement.

[0151] 3. GAT (Graph Attention Network) is a graph neural network that allows nodes to dynamically pay attention to their neighboring nodes.

[0152] 4. GCN (Graph Convolutional Network), a classic graph neural network, learns the representation of nodes by aggregating information from neighboring nodes.

[0153] 5. GAE (Graph Auto-Encoders) is an unsupervised learning model for graph structured data. It learns the low-dimensional representation (embedding) of nodes in the graph through an encoder-decoder structure.

[0154] 6. NCNC (Neural Common Neighbor with Completion) is a graph neural network model specifically used for link prediction tasks. This model improves the performance of link prediction by studying the incompleteness of the graph.

[0155] 6.4 Experimental Results

[0156] Table 4 Objective index evaluation results

[0157]

[0158] The objective indicator evaluation results shown in Table 4 show that the LPAPE model performs significantly better than other methods in the link prediction task. Specifically, LPAPE achieved the highest scores in the four key indicators of Accuracy, Auprc, Auroc, and F1 score, which are 0.930, 0.914, 0.889, and 0.922 respectively. This outstanding performance highlights the high accuracy and robustness of LPAPE in dealing with link prediction problems.

[0159] Compared with the second-best NCNC model, LPAPE improves by 28%, 16.4%, 14.9% and 24.2% in Accuracy, Auprc, Auroc and F1 score respectively. This significant improvement shows that LPAPE can effectively capture the complex relationships and dependency information in the implicit heterogeneous graph of the industrial chain when dealing with link prediction problems. These significant improvement percentages show that LPAPE can effectively capture the complex relationships and dependency information in the heterogeneous graph of the industrial chain, thereby achieving significant performance improvement in link prediction tasks.

[0160] The relatively poorly performing models GCN and GAE failed to meet expectations in some indicators. This may be related to the fact that GCN relies too much on the first-order neighbor information of nodes and ignores the high-order neighbor information, which leads to limitations in processing complex graph structures. Although GAE has achieved certain results in Auprc and Auroc, it performs poorly in Accuracy and F1 score. This may be due to the limitations of GAE in processing specific types of data. For example, it is more suitable for processing graph data with clear edge labels, while its performance is affected when processing missing or ambiguous edge labels. In addition, GAE may not fully explore the potential relationship between nodes when encoding graph structure information, which affects its performance in link prediction tasks.

[0161] Table 5 Integrity assessment results

[0162]

[0163] Table 5 presents the evaluation indicators of the integrated circuit industry chain data by seven methods. Among these data, our model (LPAPE) achieved the highest score in the industry chain integrity assessment. According to the definition of the parameters, the resistance of the industry chain is measured by the number of connections between enterprises and other enterprises. The more connections, the stronger the resistance of the enterprise. Similarly, the resilience of the industry chain is evaluated by the number of potential connections (i.e., hidden edges). The more hidden edges, the faster the enterprise can recover when facing risks.

[0164] Our experimental results clearly reveal that our model (LPAPE) has made significant progress in the three industry chain integrity evaluation indicators, namely, Resistance, Resilience, and Integrity score, which are improved by 0.05, 0.27, and 0.16 respectively compared with the second-best NCNC model. This fully demonstrates the excellent performance of LPAPE in mining the dependencies between enterprises in the industry chain. Further analysis of the reasons behind it shows that compared with NCNC, our model first uses GCN for feature extraction, and then introduces the attention mechanism and position encoding, which greatly expands the model's capture range of feature information and almost fully covers the feature space of industry chain data. Unlike NCNC, which only focuses on structural features, LPAPE not only considers the local features of industry chain data in the feature extraction process, but also takes into account the global features, and carefully pre-processes the industry chain data nodes through GCN. On this basis, the attention mechanism is combined to give different weights to different features. This strategy explains the reason for the performance improvement of our LPAPE model.

[0165] Table 6 Link prediction results of scientific literature graph dataset

[0166]

[0167] After that, we conducted link prediction experiments on the scientific literature graph dataset. As shown in Table 6, in the link prediction experiments on the scientific literature graph dataset, the LPAPE model showed excellent performance on the three datasets of Cora, Citeseer and Pubmed, with MRR values ​​of 39.50, 65.50 and 40.20 respectively, significantly ahead of other models. Taking the Citeseer dataset as an example, the MRR value of LPAPE is 15.55 higher than that of GCN (49.95), and on the Pubmed dataset, the MRR value of LPAPE is almost twice that of GCN. These significant differences reveal the significant advantages of LPAPE in identifying link relationships between nodes, which may come from the position encoding and attention mechanisms integrated in the LPAPE model. These mechanisms can fully cover potential link predictors in the feature encoding stage, thereby providing the model with deeper graph representation learning and link prediction capabilities. Therefore, the LPAPE model not only performs well in link prediction tasks, but its advanced technical strategies also provide new perspectives for the analysis and understanding of graph data.

[0168] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An industrial chain integrity assessment model, characterized in that: It includes a node feature encoding module, a candidate node set selection module, a pairwise encoding generation module and an evaluation module; The node feature encoding module is used to learn the feature representation of nodes in the implicit heterogeneous graph of the industrial chain based on the message passing neural network to obtain the feature encoding of the nodes; The candidate node set selection module is used to select nodes that have a significant impact on the target node from the implicit heterogeneous graph of the industrial chain according to the feature coding of the node and based on the PPR score to obtain a candidate node set; The paired coding generation module is used to combine the feature coding of the candidate nodes and the relative position coding of the candidate nodes relative to the target node pair, calculate the paired coding of the target node pair based on the graph attention mechanism, and capture the complex relationship between the nodes; The evaluation module is used to predict the existence probability of the target node pair through a multi-layer perceptron, a comprehensive candidate node set, feature coding of the target node, and paired coding of the target node pair, and to quantitatively evaluate the predicted existence probability through an integrity evaluation index.

2. The industrial chain integrity assessment model according to claim 1 is characterized in that: The node feature encoding module uses a graph convolutional network as an encoder to encode the features of each node in the implicit heterogeneous graph of the industrial chain; the graph convolutional network uses the ReLU function as the activation function; the graph convolutional network contains multiple graph convolutional layers, each layer will update the feature representation of the node, and can capture neighborhood information of different scales; the output of each layer becomes the input of the next layer, thereby allowing information to propagate throughout the graph structure. After learning through multi-layer message passing, node i learns the node feature encoding containing neighbor information and global information; Before encoding the implicit heterogeneous graph of the industrial chain, it needs to be preprocessed, including: representing the implicit heterogeneous graph of the industrial chain as an adjacency matrix and a node feature matrix; adding self-loops to each node in the adjacency matrix; and normalizing the adjacency matrix.

3. The industrial chain integrity assessment model according to claim 1 is characterized in that: The PPR score calculation formula used by the candidate node set selection module is: Among them, PPR(i,k) represents the personalized PageRank value from node i to node k, α is the damping factor, represents the set of neighbor nodes of node j, deg(k) represents the out-degree of node k, δ ij is an indicator function. When i=j, δ ij =1, otherwise δ ij =0.

4. The industrial chain integrity assessment model according to claim 3 is characterized in that: Assuming that there is a target node pair (a, b) in the implicit heterogeneous graph of the industrial chain, the methods for screening out the nodes that have a significant impact on it include: Calculate the personalized PageRank value ppr(a,u) from node a to node u, and the personalized PageRank value ppr(b,u) from node b to node u, where node u is the root node in the implicit heterogeneous graph of the industrial chain; For a certain type of neighbor node of the target node pair (a, b) in the implicit heterogeneous graph of the industrial chain, a PPR threshold η is set for it, and then all nodes whose PPR scores relative to the two nodes a and b in the target node pair (a, b) are retained, that is, the nodes with significant influence on the target node pair (a, b) are obtained; The specific selection implementation formula for filtering node sets of various nodes that have significant impact on the target node pair (a, b) is as follows: in, is the filtered node set of neighbor nodes of node type π, η π is the PPR threshold corresponding to the neighbor node type π.

5. The industrial chain integrity assessment model according to claim 4 is characterized in that: The paired encoding formula used by the paired encoding generation module is: Where w(a,b,u) measures the importance of node u to the target node pair (a,b), h(a,b,u) is the feature encoding of node u relative to (a,b), V is the set of all nodes in the graph, ⊙ is the Hadamard product, which means the multiplication of corresponding elements of two vectors; The feature encoding h(a,b,u) of node u relative to (a,b) is defined as: Among them, h u Table node u feature encoding, represents the relative position encoding of node u relative to the target node pair (a, b), ∥ is the connection operator, and W is a learnable weight matrix used to perform linear transformation on the concatenated vector; The importance w(a,b,u) of node u to the target node pair (a,b) is modeled by attention as follows: Among them, h a ,h b Represent the feature codes of nodes a and b respectively; represents the set of nodes remaining after removing nodes a and b in the graph; φ(·) is the GATv2 attention mechanism; is the unnormalized attention weight of node u for (a, b); w(a, b, u) is the normalized attention weight obtained after using the softmax function.

6. The industrial chain integrity assessment model according to claim 5 is characterized in that: The relative position encoding calculation formula of node u relative to the target node pair (a, b) is: rpe (a,b,u) =MLP(ppr(a,u),ppr(b,u)) Among them, MLP is a multi-layer perceptron, which is used to parameterize the relative position encoding calculation method.

7. The industrial chain integrity assessment model according to claim 6 is characterized in that: The prediction score function used by the evaluation module is: Where p(a,b) is the probability of predicting that there is a link between node a and node b; σ is the sigmoid function; ⊙ is the Hadamard product; ∥ is the join operator; h a represents the feature code of node a; s(a,b) represents the pairwise coding information formed by the node pair (a,b); and Belonging to the candidate node set, they represent the number of common neighbors, the number of one-hop neighbors, and the number of neighbors greater than one hop of the node pair (a, b).

8. The industrial chain integrity assessment model according to claim 7 is characterized in that: The calculation formula for the industry chain integrity score is as follows: Among them, RT(C) represents the resistance measurement of the industrial chain, and RL(C) represents the recovery measurement of the industrial chain.

9. An industrial chain integrity assessment system, characterized in that: The system is embedded with the industrial chain integrity assessment model described in any one of claims 1 to 8.

10. A method for evaluating the integrity of an industrial chain, characterized in that: The method evaluates the integrity of the industrial chain based on the industrial chain integrity evaluation model described in any one of claims 1 to 8.

Images