Anti-counterfeiting traceability method and system based on big data analysis

Abstract

The invention provides an anti-counterfeiting traceability method and system based on big data analysis, and the method comprises the steps: obtaining a supply chain relation network comprising a plurality of traceability vertexes, carrying out the dynamic weight distribution operation of the traceability vertexes in the supply chain relation network based on the types of direct-connection traceability vertexes, and obtaining the type dimension focusing influence coefficient of each traceability vertex; based on the category dimension focusing influence coefficient of each traceability vertex, performing vertex dimension dynamic weight distribution operation on each traceability vertex to obtain a vertex dimension focusing influence coefficient of each traceability vertex; based on the vertex dimension focusing influence coefficient of each traceability vertex, vertex feature extraction is carried out on the traceability vertexes in the supply chain relation network, a vertex coding vector of each traceability vertex is obtained, anti-counterfeiting traceability detection is carried out on each traceability vertex, and an anti-counterfeiting traceability detection result of each traceability vertex is obtained. According to the method, the reliability of forged traceability vertex detection can be improved, and meanwhile, the computing power overhead during data processing can be reduced.

Description

Technical Field

[0001] This application relates to the field of data processing, and more particularly, to an anti-counterfeiting and traceability method and system based on big data analysis. Background Art

[0002] In the context of the current prosperous development of the commodity economy, anti-counterfeiting and traceability technologies are crucial for protecting consumer rights and interests, maintaining market order, and enhancing corporate image. Existing anti-counterfeiting and traceability technologies have their limitations. Although electronic tag technology can achieve target identification and data exchange, it is vulnerable to signal interference and the cost of reading and writing devices is high. Although two-dimensional code recognition technology allows consumers to obtain product information, two-dimensional codes are easily copied and tampered with. Anti-counterfeiting labels have problems such as being reusable and not traceable. Existing anti-counterfeiting and traceability technologies have deficiencies in aspects such as anti-counterfeiting ability, traceability accuracy, cost control, and anti-copying and tampering. With the iteration of technology, the application of using a graph structure for traceability and anti-counterfeiting can be adopted. After characterizing the features of the vertices therein, anti-counterfeiting identification is performed. However, this method has insufficient reliability in vertex identification and has a large dependence on computing power, resulting in high costs. Summary of the Invention

[0003] The purpose of the present invention is to provide an anti-counterfeiting and traceability method and system based on big data analysis.

[0004] In a first aspect, this application provides an anti-counterfeiting and traceability method based on big data analysis. The method includes: obtaining a supply chain relationship network including a plurality of traceability vertices, where the types of the traceability vertices are product traceability vertices or transfer traceability vertices; performing a dynamic weight assignment operation on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices to obtain a type dimension focus influence coefficient for each traceability vertex; based on the type dimension focus influence coefficient of each traceability vertex, performing a vertex dimension dynamic weight assignment operation on each traceability vertex to obtain a vertex dimension focus influence coefficient for each traceability vertex; based on the vertex dimension focus influence coefficient of each traceability vertex, performing vertex feature extraction on the traceability vertices in the supply chain relationship network to obtain a vertex coding vector for each traceability vertex; and based on the vertex coding vector of each traceability vertex, performing anti-counterfeiting and traceability detection on each traceability vertex to obtain an anti-counterfeiting and traceability detection result for each traceability vertex.

[0005] In a second aspect, this application provides a computer system, including: one or more processors; a memory; one or more computer programs; where the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method described above is implemented.

[0006] The beneficial effects of the present application at least include: based on the dynamic weight allocation operation for the traceability vertices in the supply chain relationship network according to the types of directly connected traceability vertices, the traceability vertices can better integrate the attribute feature vectors of traceability vertices of different connection line types, and based on the vertex dimension dynamic weight allocation operation for the traceability vertices, the traceability vertices learn the attribute feature vectors of the directly connected traceability vertices, increasing the feature representation effect of the vertex coding vectors of the traceability vertices. Combining with the accurate vertex coding vectors of the traceability vertices for anti-counterfeiting traceability detection can increase the reliability of forged traceability vertex detection, and at the same time reduce the computing power overhead during data processing.

[0007] In the following description, some other features will be partly stated. When examining the following content and the drawings, those skilled in the art will partly discover these features, or may learn these features through production or application. Through practicing or using various aspects of the methods, tools, and combinations listed in the detailed examples described later, the features in the current application can be implemented and obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] To more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for the description in the embodiments of the present application will be briefly introduced below.

[0009] Figure 1 is a flowchart of an anti-counterfeiting traceability method based on big data analysis provided by an embodiment of the present application.

[0010] Figure 2 is a schematic diagram of the composition of a computer system provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] The embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation part of the embodiments of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0012] The execution subject of the anti-counterfeiting traceability method based on big data analysis in the embodiments of the present application is a computer system, including but not limited to servers, personal computers, laptops, tablets, smart phones, etc. The computer system can run independently to implement the present application, or can be connected to the network and implement the present application through interaction with other computer systems in the network. Among them, the network where the computer system is located includes but not limited to the Internet, wide area network, metropolitan area network, local area network, VPN network, etc.

[0013] The embodiments of the present application provide an anti-counterfeiting traceability method based on big data analysis, as Figure 1 shown, the method includes: Step S100: Obtain a supply chain relationship network including multiple traceability vertices, where the types of traceability vertices are product traceability vertices or transfer traceability vertices.

[0014] In step S100, obtain a supply chain relationship network including multiple traceability vertices, where the types of traceability vertices are product traceability vertices or transfer traceability vertices. The supply chain relationship network is a network structure that describes the relationships between various links in the supply chain, presented in a graph structure, where the graph vertices are the traceability vertices, and the connecting lines between the vertices represent the associations between the vertices. Product traceability vertices represent information vertices related to the product itself. For example, for a mobile phone, it can include information such as the product model, batch number, production date, etc.; transfer traceability vertices represent information vertices related to the transfer process of the product in the supply chain. Information such as logistics nodes, quality inspection records, sales channels, and liability subject bindings can all be represented by transfer traceability vertices. For example, for a mobile phone from the manufacturer to the distributor and then to the consumer, the logistics nodes such as warehouses and transport vehicles involved in this process can be represented by transfer traceability vertices.

[0015] The process of obtaining the supply chain relationship network can mainly include three stages: data collection, data preprocessing, and network construction.

[0016] In the data collection and preprocessing stage, structured and unstructured data can be extracted from data sources such as enterprise resource planning (ERP) systems, Internet of Things sensors, blockchain ledgers, and third-party logistics platforms. The data of product traceability vertices comes from the bill of materials (BOM), serial number management records, and batch coding information in the production and manufacturing links. Each product traceability vertex corresponds to an item instance with spatio-temporal uniqueness, and its attribute vector contains dimensions d p of features such as production timestamp t prod 、raw material supplier code s mat 、production equipment number e id 、quality inspection index set etc. The data of transfer traceability vertices comes from logistics track records, warehousing inbound documents, customs clearance documents, and sales order flows. Its attribute vector contains fields such as transport vehicle identifier 、starting and ending position coordinates 、environmental monitoring parameters (temperature T, humidity H, vibration intensity Vib), and liability subject signature sig entity etc.

[0017] In the vertex relationship modeling stage, establish directed connecting lines between vertices based on the item movement path and business event sequence. When product traceability vertex p i passes through a certain transfer traceability vertex t jWhen a physical location or ownership change occurs, the system creates a connection line i from p j to t , and conversely establishes a reverse connection line j from t i to p to support two-way traceability query. Each connection line in the connection line attribute matrix carries an influence coefficient , which is calculated by fusing multi-modal evidence. The specific formula is: ; where w k represents the preset weight of the k-th type of transfer situation (such as logistics timeliness, quality inspection pass rate, transaction compliance), is the normalized scoring function for this transfer situation, is the original value of the k-th type of evidence extracted from the original data. For example, in the cold chain transportation scenario, if the proportion of time when the temperature record exceeds the allowable range is , then the corresponding can be defined as , is the attenuation coefficient.

[0018] In the network topology optimization stage, the community discovery algorithm is used to identify functional modules in the supply chain. For example, the Louvain method is used to maximize the modularity index: ; where is the adjacency matrix element, is the degree of vertex i, m is the total number of connection lines, takes 1 when vertices i and j belong to the same community and 0 otherwise. This process divides the network into subgraphs such as production clusters, logistics clusters, and sales clusters, laying a foundation for subsequent local feature extraction.

[0019] In terms of the dynamic update and maintenance mechanism, an incremental graph update pipeline is constructed. When new Internet of Things events (such as GPS location updates) or transaction records arrive, vertex attribute updates and connection line weight recalculations are triggered in real time through an event-driven architecture. For time-sensitive attributes (such as the remaining shelf life of perishable items), a sliding window mechanism is implemented to adjust the influence weight of historical data according to the time decay function , where is the decay rate parameter.

[0020] Step S200: Perform a dynamic weight assignment operation on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices to obtain the type dimension focused influence coefficient of each traceability vertex.

[0021] In step S200, a dynamic weight assignment operation is performed on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices, so as to obtain the type dimension focusing influence coefficient of each traceability vertex. This step is to consider the types of directly connected traceability vertices and assign different weights to different types of directly connected traceability vertices to more accurately reflect the mutual influence between traceability vertices. In the supply chain relationship network, each traceability vertex has its directly connected traceability vertices. The directly connected traceability vertices refer to other traceability vertices directly connected to this traceability vertex, and the types of these directly connected traceability vertices may be product traceability vertices or transfer traceability vertices. The dynamic weight assignment operation is a process of adjusting weights according to the types of directly connected traceability vertices. Different types of directly connected traceability vertices may have different degrees of influence on the target traceability vertex, so weights are dynamically assigned to them.

[0022] Specifically, its calculation process integrates the graph attention mechanism and the multi-modal feature interaction technology, and the specific implementation process can be as follows: Traverse each traceability vertex in the supply chain relationship network , and extract all its directly connected neighbor vertices . Based on the vertex type set C = {product traceability vertex, transfer traceability vertex}, divide into C subsets , where represents the type identifier of vertex . For each subset , perform feature weighted aggregation based on the connection line influence coefficient to generate a type-related attribute characterization vector , and the formula is expressed as: ; where is the connection line influence coefficient calculated in step S201, is the learnable weight matrix corresponding to type c, is the original attribute feature vector of vertex v j , is a non-linear activation function (such as GELU). Through the type-specific projection matrix W c , the decoupled mapping of the cross-type feature space is realized to ensure that product and transfer type vertices follow different conversion rules during feature aggregation.

[0023] For each type , construct an attention scoring mechanism between the current vertex v i and its type subset . Define the attention function , and calculate the association strength between v i and v j under type c: ; where, is the query and key transformation matrix of type c,​ , which is the attention parameter vector, represents the vector concatenation operation. Through the Softmax function, is normalized to obtain the normalized attention weight , which reflects the importance of v j under the category c for v i .

[0024] Combine the category-related attribute representation vector with the attention weight to generate the category dimension focusing influence coefficient . Specifically, first calculate the category-level context vector: ; Subsequently, fuse and through the gating mechanism, and the formula is: ; where and are the gating parameters, quantifies the overall contribution of category c to the feature learning of vertex v i . This coefficient suppresses the influence of noise or abnormal connections by adaptively adjusting the information flow intensity of different category neighbors.

[0025] In addition, the focusing coefficients of various category dimensions can be dynamically fused with the global vertex features. The formula for category-aware feature update is, for example: ; where , which is the category feature fusion matrix, and h i ' is the fused vertex feature. In this way, while retaining the vertex's own features, category-related context information is injected to enhance the discriminability of the feature representation.

[0026] Step S300: Based on the category dimension focusing influence coefficient of each source vertex, perform a vertex dimension dynamic weight assignment operation on each source vertex to obtain the vertex dimension focusing influence coefficient of each source vertex.

[0027] In step S300, based on the category dimension focusing influence coefficient of each source vertex, perform a vertex dimension dynamic weight assignment operation on each source vertex, thereby obtaining the vertex dimension focusing influence coefficient of each source vertex. In this step, based on the obtained category dimension focusing influence coefficient, the relationship between the source vertex and its directly connected source vertices will be further considered to perform a more refined weight assignment to highlight the specific influence of different directly connected source vertices on the target source vertex.

[0028] The influence coefficient focusing on the type dimension reflects the comprehensive influence degree of different types of directly-connected traceability vertices on the target traceability vertex, while the influence coefficient focusing on the vertex dimension focuses on the interaction between a single traceability vertex and its respective directly-connected traceability vertices, aiming to more precisely capture the unique influence of each directly-connected traceability vertex on the target traceability vertex.

[0029] When performing the dynamic weight assignment operation for the vertex dimension, the weights will be adjusted based on the influence coefficient focusing on the type dimension and the specific relationship between the traceability vertex and the directly-connected traceability vertices. In this process, the characteristics of each directly-connected traceability vertex and its connection strength with the target traceability vertex are considered. To implement the dynamic weight assignment operation for the vertex dimension, various technical means can be adopted. One feasible method is the extended application of the attention mechanism. The attention mechanism can help automatically learn the importance of each directly-connected traceability vertex, thereby dynamically assigning weights. Specifically, an attention model can be constructed. The input of this model includes the attribute feature vector of the target traceability vertex, the attribute feature vector of the directly-connected traceability vertex, and the influence coefficient focusing on the type dimension. The output of the model is the weight of each directly-connected traceability vertex.

[0030] Through step S300, based on the influence coefficient focusing on the type dimension generated in step S200, the dynamic weight assignment for the vertex dimension is further implemented, thereby quantifying the fine-grained association strength between each traceability vertex and its directly-connected neighbors in the supply chain relationship network, and finally generating the influence coefficient focusing on the vertex dimension. This coefficient captures the conduction effect of the key path in the complex supply chain topology by integrating cross-type feature interaction and vertex-level attention mechanism. As a specific implementation method, each traceability vertex in the supply chain relationship network can be traversed , first load the set of influence coefficients focusing on the type dimension calculated for it in step S200 , where C = {product traceability vertex, transfer traceability vertex}, which is the set of vertex types. For each directly-connected neighbor vertex , extract the corresponding according to its type identifier , and inject this coefficient as the initial weight into the relationship modeling process between vertices. The core of the dynamic weight assignment for the vertex dimension lies in constructing a differentiable attention scoring function, which simultaneously considers vertex attribute similarity and type-related influence strength, and can be expressed as: , where is the transformation matrix of the query and the key, is the feature vector of vertex v i and v j , is the learnable parameter vector, represents the vector concatenation operation, is a tuning hyperparameter for the influence strength of categories. In this way, the category dimension focusing coefficient can correct the pure data-driven weight assignment in the traditional attention mechanism and explicitly introduce business rule constraints. Subsequently, the original attention scores are subjected to degree-based normalization to eliminate the bias caused by network heterogeneity (such as the dominance effect of high-degree vertices): ; where is the original attention score between a vertex and its directly connected neighbor vertices, is the original attention score between a vertex and any one of its directly connected neighbor vertices v k ; are the degrees of vertices v i and v j respectively, enhances the robustness of the model to the long-tail distribution by suppressing the spurious associations between high-degree vertices. The normalized attention weights constitute the main part of the vertex dimension focusing influence coefficient. However, to further improve the model's perception ability of the supply chain hierarchical structure, a dynamic routing mechanism is introduced to iteratively adjust the weight distribution of the information propagation path. Specifically, the dynamic routing mechanism iteratively optimizes the routing coupling coefficient , and directly affects the starting point of the routing process as the initial weight. Specifically: at the first iteration (t = 0), the routing coupling coefficient can be initialized as a linear transformation of : ; where W b and b b are learnable parameters that map to the routing space. This design makes the local attention information carried by the basis for routing optimization. The routing coupling coefficient is updated in the t-th iteration as: ; where is the routing transformation matrix at the t-th iteration, is the routing parameter vector. After T iterations, the final vertex dimension focusing influence coefficient is generated through the Softmax function: .

[0031] This coefficient integrates the initial attention weights and the structural prior knowledge learned in the dynamic routing process, and can adaptively identify the key conduction nodes in the supply chain (such as the core quality inspection link or the main logistics hub). To enhance the model's ability to express the polysemy of vertex attributes, a multi-head attention expansion mechanism is adopted to decompose the feature space into K independent subspaces, and each subspace performs an independent weight assignment process: ; where represents concatenation along the feature dimension, is the value transformation matrix of the kth attention head, The vertex dimension-focused influence coefficient calculated for the kth head. This multi-head design allows the model to capture differentiated association patterns in different semantic subspaces, such as strengthening time series correlation in one head and focusing on geographic proximity in another head. Residual connections and layer normalization (LayerNorm) are further introduced to stabilize the training process and prevent gradient disappearance or explosion: ;in is the output transformation matrix, which is used to map the concatenated multi-head features back to the original dimension. In order to optimize the discriminative ability of vertex dimension focusing influence coefficient, a contrast loss function is designed to force the model to shorten the feature distance of positive sample vertex pairs (with real supply chain associations) and push away the similarity of negative sample pairs (randomly sampled unrelated vertices): ;in is a cosine similarity function, and P and N represent the sets of positive and negative sample pairs, respectively. This loss function is jointly optimized with the category dimension contrast loss in step S200 to form a multi-level feature constraint system.

[0032] Step S400: Based on the vertex dimension focusing influence coefficient of each traceability vertex, vertex features are extracted from the traceability vertices in the supply chain relationship network to obtain a vertex encoding vector of each traceability vertex.

[0033] In step S400, based on the vertex dimension focusing influence coefficient of each traceability vertex, vertex features of the traceability vertex in the supply chain relationship network are extracted to obtain the vertex encoding vector of each traceability vertex. In the entire anti-counterfeiting traceability method process, after the processing of the previous steps, the vertex dimension focusing influence coefficient of each traceability vertex has been obtained, which reflects the important degree of association between each traceability vertex and its directly connected traceability vertex. Step S400 uses these coefficients to extract a vector that can accurately represent the characteristics of each traceability vertex from the supply chain relationship network, that is, a vertex encoding vector, so as to perform subsequent anti-counterfeiting traceability detection.

[0034] Step S400 can achieve deep abstraction and key pattern extraction of vertex attributes in the supply chain network based on the multi-order message passing framework of the graph neural network, combined with dynamic routing optimization and residual feature learning. First, load the vertex dimension focus influence coefficient generated in step S300 , where |V| is the total number of vertices, Represents vertex v i With directly connected neighbors v j The core of vertex feature extraction is to design a differentiable feature aggregation function , which converts the vertex attributes Weighted features with its neighbors Map to a high-dimensional semantic space. In specific implementation, a hybrid architecture of multi-head graph attention mechanism and residual gated network can be adopted: ; Among them, is the feature vector of vertex v i at the l+1-th layer; is the input feature vector of vertex v i at the l-th layer; l represents the network layer index, K is the number of attention heads, and are the value transformation matrix and output projection matrix of the k-th head respectively, is the vertex dimension focusing influence coefficient independently calculated by the k-th head. This architecture captures diverse association patterns in the supply chain (such as logistics path dependence, quality inspection compliance association, etc.) through multi-head decomposition, while residual connections retain the original feature information to prevent over-smoothing. To further enhance the model's ability to model long-range dependencies and non-local interactions, a dynamic routing protocol is introduced to iteratively optimize the weight distribution of the feature propagation path. In the t-th routing iteration, the hidden state i of vertex v is updated as: ; among them is the gated recurrent unit, is the hidden state vector of vertex v i after the (t-1)-th routing iteration, is the routing transformation matrix, is the dynamic weight of vertex v i and v j in the t-th routing iteration, is dynamically adjusted according to the current hidden state: ; among them is the query and key transformation matrix, is the learnable parameter vector. After T iterations, the final vertex encoding vector fuses the semantic information of multi-order neighbors and the structural prior of dynamic routing optimization. To improve the robustness of feature representation, a feature decoupling strategy is implemented, and the vertex encoding vector is decomposed into a topology-related component and an attribute-related component : ; among them is the decoupling projection matrix. In this way, the model can optimize the representation learning of network topology features and vertex inherent attributes respectively, and enforce the independence of the two components through the orthogonal regularization term .

[0035] Step S500: Based on the vertex encoding vectors of each traceability vertex, perform anti-counterfeiting traceability detection on each traceability vertex to obtain the anti-counterfeiting traceability detection results of each traceability vertex.

[0036] To implement anti-counterfeiting traceability detection, machine learning or deep learning methods can be used. A feasible way is to build a classification model with the vertex encoding vector as the input, and the output of the model is the judgment result on whether the traceability vertex is forged. Classification models include logistic regression, support vector machines, neural networks, etc. Taking neural networks as an example, a multi-layer perceptron (MLP) model can be built. Its input layer receives the vertex encoding vector, the hidden layer performs non-linear transformation and feature extraction on the input features, and the output layer outputs the probability that the traceability vertex is forged through an activation function (such as the Sigmoid function).

[0037] For example, first load the vertex encoding vectors generated in step S400 and their decoupled components (topological features) and (attribute features), and at the same time load the original attribute feature vector of the vertex , construct a combined feature vector through a cross-modal feature interaction module , ; where is the feature fusion matrix, is the bias term, represents the vector concatenation operation, is the Gaussian error linear unit activation function. In this way, the high-dimensional encoded features and the original attribute features are mapped to a unified hidden space d h through non-linear mapping, enhancing the model's ability to capture implicit association patterns. The forgery probability calculation module uses a hierarchical fully connected network, and its output layer is: ; where, and are the weight matrices of the hidden layer and the output layer, and are the bias terms, is the Sigmoid function, , indicating that vertex v i is the probability estimate value of a forged vertex.

[0038] As an implementation, before step S200, which performs dynamic weight allocation operations on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices to obtain the type dimension focusing influence coefficients of each traceability vertex, the method further includes: Step S201: Based on one or more transfer situations between the product traceability vertex and the transfer traceability vertex, determine the influence coefficient of the connection line between the product traceability vertex and the transfer traceability vertex.

[0039] In step S201, based on one or more transfer situations between the product traceability vertex and the transfer traceability vertex, determine the influence coefficient of the connection line between the product traceability vertex and the transfer traceability vertex. The product traceability vertex represents the information vertex related to the product itself, while the transfer traceability vertex represents the information vertex related to the product during its transfer in the supply chain. The influence coefficient of the connection line reflects the degree of mutual influence between the product traceability vertex and the transfer traceability vertex under different transfer situations.

[0040] The transfer situations cover all aspects of the product from production to sales in the supply chain, including logistics node transfer, quality inspection records, sales channels, liability entity binding, etc. Different transfer situations have different impacts on the association between the product traceability vertex and the transfer traceability vertex.

[0041] The influence coefficient of the connection line is a quantitative indicator used to measure the degree of mutual influence between the product traceability vertex and the transfer traceability vertex. The larger the influence coefficient, the closer the association between the two and the greater the mutual influence; conversely, the smaller the influence coefficient, the weaker the association between the two and the smaller the mutual influence.

[0042] To determine the influence coefficient of the connection line, a feasible method is based on data mining and machine learning. First, collect data on various transfer situations between the product traceability vertex and the transfer traceability vertex, and then preprocess this data, including operations such as data cleaning and feature extraction. Next, machine learning algorithms such as regression analysis, decision trees, and neural networks can be used to establish a mapping relationship between the transfer situation and the influence coefficient of the connection line. For example, using the linear regression algorithm, the following formula can be established: ; where represents the influence coefficient of the connection line, respectively represent the feature values of transfer situations such as logistics node transfer, quality inspection records, sales channels, liability entity binding, etc., are the corresponding weight coefficients, and b is the bias term. By training the data to learn the values of these weight coefficients and bias terms, the model can accurately predict the influence coefficient of the connection line based on the transfer situation. Another method is based on expert experience and rules. Experts in the supply chain field can be invited to formulate a series of rules to determine the influence coefficient of the connection line according to their experience and knowledge. For example, experts can stipulate that if the storage time of a product in a certain warehouse exceeds a certain number of days and the quality inspection passing rate reaches a certain percentage, then the influence coefficient of the connection line between this warehouse and the product traceability vertex is a relatively high value; if the sales channels are mainly some small retailers and the credit rating of the liability entity is relatively low, then the influence coefficient is a relatively low value. Evaluate the connection line between each product traceability vertex and the transfer traceability vertex according to these rules to determine its influence coefficient.

[0043] Alternatively, it is also possible to perform an exact mapping from the original business data to the numerical relationship weights based on the feature fusion and rule reasoning of multi-source heterogeneous data, combined with business constraints and statistical learning. Specifically, all product traceability vertices can be first extracted from the supply chain relationship network and the transfer traceability vertices between the connection link sets , and for each connection link , parse the associated transfer situation set , where K represents the number of transfer situation types (such as logistics nodes, quality inspection records, sales channels, etc.).

[0044] The influence coefficient k of each transfer situation r is calculated through the following steps: For the transfer situation of logistics nodes, extract the spatio-temporal consistency index and the path compliance index from the Internet of Things sensor logs and GPS trajectory data, and its calculation formula is: ; where is the planned arrival time, is the actual arrival time, is the time tolerance threshold (such as 24 hours), l i is the coordinate of the i-th trajectory point, is the geographical fence area of the preset compliance path, N is the total number of trajectory points, is the indicator function. The influence coefficient of logistics node transfer is then obtained through weighted fusion: ; where, is the weight coefficient, satisfying , and is calculated and determined according to historical data by the entropy weight method. For the transfer situation of quality inspection records, extract the quality inspection item compliance index (0 means not passed, 1 means passed) and the environmental parameter stability index from the blockchain deposit, and the latter is defined as: ; where are the mean and variance of the m-th environmental parameter (such as temperature, humidity) during the transfer cycle respectively, is the smoothing factor (such as 10 -6 ), and M is the total number of parameters. The influence coefficient of quality inspection records is calculated through the logistic regression model: ; where are the model parameters, is the Sigmoid function. The credibility score of the sales channel Based on historical transaction records and third-party credit rating data, it is calculated through the following rules: If the sales channel is officially authorized, then ; If it is a third-party platform, then , where is the platform credit score; If it is an unknown channel, then . The influence coefficient directly takes the value of . After obtaining the influence coefficients for all K types of transfer situations, a dynamic fusion method based on the attention mechanism is used to generate the final connection line influence coefficient : ; where is the learnable attention vector for the k-th transfer situation, is the context feature vector of the connection line , which is obtained by projecting after concatenating the vertex attributes and the connection line attributes: ; where and are the attribute feature vectors of the product traceability vertex v p and the transfer traceability vertex v t respectively, is the original attribute of the connection line (such as transfer time, operator ID, etc.), and are the projection parameters. This dynamic fusion mechanism allows the model to adaptively adjust the contribution weights of different transfer situations according to specific business scenarios. For example, in cold chain logistics, the weight of the temperature stability index is increased, while in high-value commodities, the role of the sales channel credibility is strengthened. To verify the rationality of the influence coefficient calculation, a double-check mechanism is implemented: One is the supervised check based on historical abnormal events. Using the connection data of the connection lines in known forgery cases, calculate the distribution deviation in these cases. The formula is: ; where is the set of abnormal connection lines, is the preset abnormal threshold (such as 0.3). If ( is the acceptable deviation, such as 0.1), it is considered that the influence coefficient calculation is valid; The other is the logical check based on business rules.

[0045] Alternatively, as another implementation, in step S201, based on one or more transfer situations between the product traceability vertex and the transfer traceability vertex, determining the influence coefficient of the connection line between the product traceability vertex and the transfer traceability vertex includes: obtaining the influence coefficients corresponding to one or more transfer situations between the product traceability vertex and the transfer traceability vertex respectively; performing a fusion operation on the influence coefficients corresponding to one or more transfer situations between the product traceability vertex and the transfer traceability vertex respectively to obtain the influence coefficient of the connection line between the product traceability vertex and the transfer traceability vertex.

[0046] Since the transfer of products in the supply chain is a complex process involving multiple transfer situations, and the influence coefficient of each transfer situation can only reflect its unilateral effect, in order to comprehensively and accurately measure the degree of association between the product traceability vertex and the transfer traceability vertex, these influence coefficients are fused. Taking smartphones and component suppliers as an example, the influence coefficients corresponding to the four transfer situations of logistics node transfer, quality inspection records, sales channels, and responsible entity binding have been obtained as 0.8, 0.7, 0.6, and 0.7 respectively. Next, these influence coefficients are fused to obtain the final influence coefficient of the connection line.

[0047] A feasible fusion method is the weighted average method. Assign corresponding weights to each transfer situation according to its relative importance, then multiply the influence coefficient of each transfer situation by its corresponding weight, and finally add these products to obtain the influence coefficient of the connection line. In addition to the weighted average method, other fusion methods can also be used, such as the fusion method based on neural networks. A neural network model can be constructed, taking the influence coefficients of various transfer situations as inputs, and through training the neural network, allowing it to automatically learn the complex relationships between different influence coefficients, so as to output the final influence coefficient of the connection line.

[0048] As an implementation, in step S200, performing a dynamic weight assignment operation on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices to obtain the type - dimension focusing influence coefficient of each traceability vertex, including: Performing the following operations on the first traceability vertex in the supply chain relationship network: Step S210: Determine one or more directly connected traceability vertices of any type of the first traceability vertex, where one or more directly connected traceability vertices of any type are traceability vertices connected to the first traceability vertex based on any type of transfer situation; Step S220: Performing an integration operation on the traceability vertex attribute feature vectors of one or more directly connected traceability vertices to obtain the attribute representation vector of the first traceability vertex under any type; Step S230: Perform a dynamic weight allocation operation on the attribute representation vector of the first traceability vertex under any category and the traceability vertex attribute feature vector of the first traceability vertex to obtain the focus influence coefficient of the first traceability vertex under any category; Step S240: Use the set of focus influence coefficients of the first traceability vertex under multiple categories as the category dimension focus influence coefficient of each traceability vertex.

[0049] In step S210, determine one or more directly connected traceability vertices of any category of the first traceability vertex. Here, one or more directly connected traceability vertices of any category are the traceability vertices connected to the first traceability vertex based on the transfer situation of any category. As mentioned before, in the supply chain relationship network, each traceability vertex has its directly connected traceability vertices, and these directly connected traceability vertices are connected to the first traceability vertex through different types of transfer situations, such as logistics node transfer, quality inspection records, sales channels, and responsible entity binding.

[0050] In step S220, perform an integration operation on the traceability vertex attribute feature vectors of one or more directly connected traceability vertices to obtain the attribute representation vector of the first traceability vertex under any category. The traceability vertex attribute feature vector is a feature expression of the basic information of the traceability vertex and contains various attribute information of the traceability vertex. For example, for the transfer traceability vertex corresponding to a logistics company, its attribute feature vector may contain information such as transportation capacity, transportation cost, and transportation time; for the transfer traceability vertex corresponding to an automobile dealer, its attribute feature vector may contain information such as sales performance, market coverage, and customer evaluation.

[0051] In the integration operation, first determine the influence coefficient of the connection line between each directly connected traceability vertex and the first traceability vertex. This influence coefficient reflects the influence degree of the directly connected traceability vertex on the first traceability vertex, and its determination method has been elaborated in detail in the previous step S201. For example, it can be obtained by fusing the influence coefficients corresponding to transfer situations such as logistics node transfer, quality inspection records, sales channels, and responsible entity binding. Then, based on the influence coefficient of the connection line between each directly connected traceability vertex and the first traceability vertex, perform a fusion operation on the traceability vertex attribute feature vectors of each directly connected traceability vertex to obtain the attribute representation vector of the first traceability vertex under any category, such as weighted summation.

[0052] In step S230, a dynamic weight allocation operation is performed on the attribute representation vector of the first traceability vertex under any category and the traceability vertex attribute feature vector of the first traceability vertex to obtain the focusing influence coefficient of the first traceability vertex under any category. This step aims to comprehensively consider the attribute characteristics of the first traceability vertex itself and the attribute representations of its directly connected traceability vertices under a certain category, and determine the focusing influence degree of this category on the first traceability vertex through dynamic weight allocation. First, the attribute representation vector of the first traceability vertex under any category and the traceability vertex attribute feature vector of the first traceability vertex are concatenated to obtain the attribute concatenation vector of the first traceability vertex under any category. Suppose the traceability vertex attribute feature vector of the first traceability vertex is a, and the attribute representation vector under a certain category is b, then the attribute concatenation vector c is the vector obtained by connecting a and b in sequence, for example, c = [a; b] (here the semicolon represents the vector concatenation operation).

[0053] Next, based on the adaptive variable of any category, a feature transformation is performed on the attribute concatenation vector of the first traceability vertex under any category to obtain the focusing influence coefficient of the first traceability vertex under any category. The adaptive variable is a learnable parameter, such as the weight matrix of an attention network. Through these parameters, a linear transformation and a non-linear mapping can be performed on the attribute concatenation vector, so as to explore the potential relationships between different features. Suppose the adaptive variables are the weight matrix W and the bias vector b0, and the activation function is (such as the Sigmoid function), then the focusing influence coefficient z can be calculated by the following formula: , Finally, in order to make the focusing influence coefficient comparable and interpretable, a normalization operation is performed on the focusing influence coefficient of the first traceability vertex under any category. For example, the Softmax function is used to convert it into a probability distribution. Suppose the focusing influence coefficients of the first traceability vertex under m categories are , then the focusing influence coefficient after being normalized by the Softmax function can be calculated by the following formula: .

[0054] In step S240, the set of the focusing influence coefficients of the first traceability vertex under multiple categories is used as the category dimension focusing influence coefficient of each traceability vertex. In the supply chain relationship network, the first traceability vertex may be connected to directly connected traceability vertices of multiple different categories, and each category will have a different degree of influence on the first traceability vertex. Through the previous steps, the focusing influence coefficients of the first traceability vertex under each category have been obtained. Combining these focusing influence coefficients forms the category dimension focusing influence coefficient of this traceability vertex.

[0055] In the actual implementation process, deep learning frameworks (such as TensorFlow or PyTorch) can be used to implement the above steps. For determining the directly connected traceability vertices, the query function of the graph database and data processing libraries (such as pandas) can be used to screen and organize the data. For the integration operation of the attribute feature vectors and the dynamic weight assignment operation, a neural network model can be constructed. By defining corresponding layers (such as fully connected layers, activation layers) and loss functions, the model can be trained using the backpropagation algorithm to learn the values of the adaptive variables. For the normalization operation, the Softmax function provided in the deep learning framework can be directly used.

[0056] By operating on the first traceability vertex in the supply chain relationship network through steps S210 - S240, the category dimension focus influence coefficient of each traceability vertex can be accurately obtained. This process fully considers the categories and attribute features of the directly connected traceability vertices. Through dynamic weight assignment and feature transformation, the traceability vertices can better integrate the information of different types of directly connected traceability vertices, providing richer and more accurate feature information for subsequent vertex dimension dynamic weight assignment and anti-counterfeiting traceability detection, which helps to improve the performance and reliability of the entire anti-counterfeiting traceability system.

[0057] As an implementation method, in step S220, an integration operation is performed on the traceability vertex attribute feature vectors of one or more directly connected traceability vertices to obtain the attribute representation vector of the first traceability vertex under any category, including: Step S221: Determine the influence coefficient of the connection line between each directly connected traceability vertex and the first traceability vertex; Step S222: Based on the influence coefficient of the connection line between each directly connected traceability vertex and the first traceability vertex, perform a fusion operation on the traceability vertex attribute feature vectors of each directly connected traceability vertex to obtain the attribute representation vector of the first traceability vertex under any category.

[0058] In step S221, the influence coefficient of the connection line between each directly connected traceability vertex and the first traceability vertex is determined. In the supply chain relationship network, the first traceability vertex and the directly connected traceability vertices are connected through different types of transfer situations. The influence coefficient of the connection line reflects the influence degree of the directly connected traceability vertex on the first traceability vertex, which comprehensively considers the effects of various transfer situations on the relationship between the two.

[0059] The method for determining the connection line influence coefficient can be based on the analysis of different transfer situations in the previous steps. In step S201, it has been mentioned that the influence coefficient corresponding to one or more transfer situations between the product traceability vertex and the transfer traceability vertex can be obtained, and these influence coefficients can be fused to obtain the influence coefficient of the connection line.

[0060] In step S222, based on the influence coefficients of the connection lines between each directly-connected traceable vertex and the first traceable vertex, a fusion operation is performed on the traceable vertex attribute feature vectors of each directly-connected traceable vertex to obtain the attribute representation vector of the first traceable vertex under any category. The traceable vertex attribute feature vector is a feature expression of the basic information of the traceable vertex, which contains various attribute information of the traceable vertex. In the fusion operation, the attribute feature vectors of the directly-connected traceable vertices are weighted according to the connection line influence coefficients to highlight the features of the directly-connected traceable vertices that have a greater influence on the first traceable vertex. Suppose the first traceable vertex has n directly-connected traceable vertices, and the attribute feature vector of the i-th directly-connected traceable vertex is , and the influence coefficient of the connection line between it and the first traceable vertex is . Then, the attribute representation vector y of the first traceable vertex under this category can be obtained by the method of weighted summation, and the formula is: .

[0061] In practical applications, in order to ensure the accuracy and effectiveness of the fusion operation, preprocessing is also performed on the attribute feature vectors. For example, different attribute features are normalized to make them have the same scale range, avoiding some features from dominating in the fusion process due to a large numerical range. The Min-Max normalization method can be used to convert each feature value x in the attribute feature vector into the normalized feature value x', and the formula is: ; where min(x) and max(x) are the minimum and maximum values of this feature among all directly-connected traceable vertices respectively.

[0062] In addition, other fusion methods can also be adopted, such as the fusion method based on neural networks. A neural network model can be constructed, taking the attribute feature vectors of the directly-connected traceable vertices and the connection line influence coefficients as inputs. Through the learning ability of the neural network, the complex relationships between different features are automatically mined, so as to obtain a more accurate attribute representation vector. In this method, the structure of the neural network can include an input layer, a hidden layer, and an output layer. The input layer receives the attribute feature vectors and influence coefficients, the hidden layer performs non-linear transformation and feature extraction on the inputs, and the output layer outputs the attribute representation vector of the first traceable vertex under this category.

[0063] During the implementation process, data analysis and machine learning libraries (such as NumPy, pandas, and scikit-learn in Python) can be used to perform preprocessing and fusion operations on the attribute feature vectors. For neural network-based fusion methods, deep learning frameworks (such as TensorFlow or PyTorch) can be used to build and train neural network models. Through these technical means, the influence coefficient of the connection line can be accurately determined, and the attribute feature vectors of the directly connected traceability vertices can be effectively fused to obtain the attribute representation vector of the first traceability vertex under any category, providing more reliable feature information for subsequent dynamic weight assignment and anti-counterfeiting traceability detection.

[0064] As an implementation manner, in step S230, a dynamic weight assignment operation is performed on the attribute representation vector of the first traceability vertex under any category and the traceability vertex attribute feature vector of the first traceability vertex to obtain the focusing influence coefficient of the first traceability vertex under any category, including: Step S231: Concatenate the attribute representation vector of the first traceability vertex under any category and the traceability vertex attribute feature vector of the first traceability vertex to obtain the attribute concatenation vector of the first traceability vertex under any category; Step S232: Based on the adaptive variable of any category, perform feature transformation on the attribute concatenation vector of the first traceability vertex under any category to obtain the focusing influence coefficient of the first traceability vertex under any category; Step S233: Perform a normalization operation on the focusing influence coefficient of the first traceability vertex under any category to obtain the focusing influence coefficient of the first traceability vertex under any category.

[0065] In step S230, a dynamic weight assignment operation is performed on the attribute representation vector of the first traceability vertex under any category and the traceability vertex attribute feature vector of the first traceability vertex to obtain the focusing influence coefficient of the first traceability vertex under this category, and steps S231 - S233 are the specific processes to implement this operation. The purpose of these steps is to comprehensively consider the characteristics of the first traceability vertex itself and the comprehensive characteristics of its directly connected traceability vertices under a specific category, and through dynamic weight assignment and feature transformation, accurately measure the focusing influence degree of this category on the first traceability vertex.

[0066] In step S231, the attribute representation vector of the first traceability vertex in any category and the traceability vertex attribute feature vector of the first traceability vertex are spliced ​​to obtain the attribute splicing vector of the first traceability vertex in the category. In the supply chain relationship network, the traceability vertex attribute feature vector of the first traceability vertex contains the basic information of the vertex itself, such as the model, specification, production batch, etc. of the product; and the attribute representation vector of the first traceability vertex in any category is obtained by integrating the attribute feature vectors of its directly connected traceability vertices, reflecting the comprehensive influence of the directly connected traceability vertices in the category on the first traceability vertex.

[0067] In step S232, based on any type of adaptive variables, the attribute splicing vector of the first traceability vertex in the type is feature transformed to obtain the focus influence coefficient of the first traceability vertex in the type. Adaptive variables are learnable parameters, such as the weight matrix of the attention network, which can be automatically adjusted according to the input attribute splicing vector, so as to mine the potential relationship between different features and realize the dynamic weight allocation of the attribute splicing vector.

[0068] In step S233, the focusing influence coefficient of the first traceability vertex in the category is standardized to obtain the focusing influence coefficient of the first traceability vertex in the category. The purpose of the standardization operation is to convert the focusing influence coefficient into a comparable and interpretable numerical value, usually converted into a probability distribution so that the sum of all focusing influence coefficients is 1.

[0069] A feasible normalization method is to use the Softmax function. Assume that the focusing influence coefficients of the first traceable vertex in m categories are , the focusing influence coefficient after being standardized by the Softmax function It can be calculated by the following formula: , where exp(x) represents an exponential function with the natural constant e as the base. Through the Softmax function, each focus influence coefficient is converted into a probability value between 0 and 1, and the sum of all probability values ​​is 1. This means that the standardized focus influence coefficient can represent the relative importance of the category to the first traceability vertex among all categories.

[0070] In the actual implementation process, the Softmax function can be directly used in the deep learning framework to standardize the focus influence coefficient. Through the standardized focus influence coefficient, we can more clearly understand the relative importance of different types to the first traceability vertex, and provide a more accurate basis for the subsequent dynamic weight allocation of vertex dimensions and anti-counterfeiting traceability detection.

[0071] By processing the attribute characterization vector and the traceability vertex attribute feature vector of the first traceability vertex under any category through steps S231 - S233, dynamic weight assignment and feature transformation are achieved, and a comparable and interpretable focus influence coefficient is obtained through standardization operations. This series of operations helps to more accurately measure the focus influence degree of different categories on the first traceability vertex, provides an important decision-making basis for subsequent anti-counterfeiting traceability detection and supply chain management, and improves the accuracy and reliability of the entire anti-counterfeiting traceability system.

[0072] As an implementation manner, step S300, based on the category dimension focus influence coefficient of each traceability vertex, performs a vertex dimension dynamic weight assignment operation on each traceability vertex to obtain the vertex dimension focus influence coefficient of each traceability vertex, including: Perform the following operations on the first traceability vertex in the supply chain relationship network: Step S310: Determine one or more directly connected traceability vertices of any category of the first traceability vertex, where one or more directly connected traceability vertices of any category are traceability vertices connected to the first traceability vertex based on the transfer situation of any category; Step S320: Obtain the focus influence coefficient of the first traceability vertex under any category from the category dimension focus influence coefficient of the first traceability vertex; Step S330: Based on the focus influence coefficient of the first traceability vertex under any category and one or more directly connected traceability vertices, perform a dynamic weight assignment operation on the first traceability vertex to obtain the vertex dimension focus influence coefficient of the first traceability vertex.

[0073] In step S310, determine one or more directly connected traceability vertices of any category of the first traceability vertex. Here, one or more directly connected traceability vertices of any category are traceability vertices connected to the first traceability vertex based on the transfer situation of this category. For example, for a product traceability vertex A, its directly connected transfer traceability vertices include, for example, vertex B related to logistics node transfer, vertex C related to quality inspection records, and vertex D related to sales channels. These directly connected traceability vertices can be determined by traversing the connection information of the supply chain relationship network. Graph traversal algorithms such as breadth-first search (BFS) or depth-first search (DFS) can be used to start from the first traceability vertex and find other traceability vertices connected to it through specific category transfer situations.

[0074] In step S320, from the category - dimension focusing influence coefficients of the first traceability vertex, obtain the focusing influence coefficient of the first traceability vertex under any one of these categories. The category - dimension focusing influence coefficients are obtained in step S200 through a dynamic weight - assignment operation on the traceability vertices in the supply - chain relationship network based on the categories of the directly - connected traceability vertices. For example, for the product traceability vertex A, under the category of logistics - node transfer, its focusing influence coefficient reflects the influence degree of the directly - connected traceability vertices of this category on the product traceability vertex A. A data structure, such as a dictionary, can be established to associatively store the traceability vertices and their focusing influence coefficients under different categories. In this way, when it is necessary to obtain the focusing influence coefficient of a certain traceability vertex under a certain category, it can be quickly searched through the identifier of the traceability vertex and the category information.

[0075] In step S330, based on the focusing influence coefficient of the first traceability vertex under any one of these categories and one or more directly - connected traceability vertices, perform a dynamic weight - assignment operation on the first traceability vertex, so as to obtain the vertex - dimension focusing influence coefficient of the first traceability vertex. This process enables the first traceability vertex to better learn the traceability - vertex attribute feature vectors of its directly - connected traceability vertices and enhances the feature - representation effect of its vertex - encoding vector.

[0076] The following further illustrates with a specific example. Assume that the product traceability vertex A has three directly - connected transfer traceability vertices B, C, and D, corresponding to three transfer situations: logistics - node transfer, quality - inspection record, and sales channel. In step S310, these three directly - connected traceability vertices are determined through a graph - traversal algorithm; in step S320, the focusing influence coefficients of the product traceability vertex A under the three categories of logistics - node transfer, quality - inspection record, and sales channel are obtained as α, β, and γ respectively from the pre - stored data structure of category - dimension focusing influence coefficients.

[0077] In the specific operation of step S330, first, splice the attribute - representation vector of the first traceability vertex and the traceability - vertex attribute feature vectors of each of the directly - connected traceability vertices to obtain an attribute - splicing vector between the first traceability vertex and each of the directly - connected traceability vertices. The attribute - representation vector is obtained in step S220 through an integration operation on the traceability - vertex attribute feature vectors of the directly - connected traceability vertices, and it reflects the comprehensive attribute characteristics of the first traceability vertex under a certain category. For example, for the product traceability vertex A and the directly - connected traceability vertex B related to logistics - node transfer, splice the attribute - representation vector of the product traceability vertex A under the category of logistics - node transfer and the traceability - vertex attribute feature vector of the directly - connected traceability vertex B to obtain a new vector, which contains the information of both.

[0078] Next, based on the focusing influence coefficient of the first traceability vertex under any one category, the attribute splicing vector between the first traceability vertex and each of the directly connected traceability vertices is weighted and adjusted to obtain an adjusted attribute splicing vector. Taking the product traceability vertex A and the directly connected traceability vertex B as an example, the attribute splicing vector is multiplied by the focusing influence coefficient α, so that under the category of logistics node transfer, the influence degree of the directly connected traceability vertex B on the product traceability vertex A can be reflected in the adjusted attribute splicing vector. The purpose of this is to highlight the importance difference of directly connected traceability vertices of different categories to the first traceability vertex. Then, a feature transformation is performed on the adjusted attribute splicing vector to obtain the focusing influence coefficient between the first traceability vertex and each of the directly connected traceability vertices. The feature transformation can be performed in a linear transformation or a non-linear transformation manner. For example, a linear transformation is performed using a fully connected layer, and its formula can be expressed as: , where is the focusing influence coefficient between the first traceability vertex i and the directly connected traceability vertex j, is the adjusted attribute splicing vector, W is the weight matrix, and b is the bias term. Through this transformation, the adjusted attribute splicing vector is mapped to a new feature space to obtain the focusing influence coefficient. Finally, a normalization operation is performed on the focusing influence coefficient between the first traceability vertex and each of the directly connected traceability vertices to obtain the vertex dimension focusing influence coefficient between the first traceability vertex and each of the directly connected traceability vertices. The normalization operation can adopt common normalization methods, such as the Softmax function, and its formula is: , where is the normalized vertex dimension focusing influence coefficient, is the unnormalized focusing influence coefficient, and the summation is performed for all directly connected traceability vertices of the first traceability vertex. Through the normalization operation, the sum of the influence coefficients of all directly connected traceability vertices on the first traceability vertex is 1, which is convenient for subsequent calculations and analyses.

[0079] As an implementation manner, in step S330, based on the focusing influence coefficient of the first traceability vertex under any one category and one or more directly connected traceability vertices, a dynamic weight allocation operation is performed on the first traceability vertex to obtain the vertex dimension focusing influence coefficient of the first traceability vertex, including: Step S331: Splice the attribute representation vector of the first traceability vertex and the traceability vertex attribute feature vector of each directly connected traceability vertex to obtain an attribute splicing vector between the first traceability vertex and each directly connected traceability vertex; Step S332: Based on the focusing influence coefficient of the first traceability vertex under any one category, weight and adjust the attribute splicing vector between the first traceability vertex and each directly connected traceability vertex to obtain an adjusted attribute splicing vector; Step S333: Perform feature transformation on the adjusted attribute concatenation vector to obtain the focusing influence coefficients between the first traceability vertex and each directly connected traceability vertex; Step S334: Perform a standardization operation on the focusing influence coefficients between the first traceability vertex and each directly connected traceability vertex to obtain the vertex dimension focusing influence coefficients between the first traceability vertex and each directly connected traceability vertex.

[0080] In step S331, the attribute representation vector of the first traceability vertex and the traceability vertex attribute feature vectors of each directly connected traceability vertex are concatenated to obtain the attribute concatenation vector between the first traceability vertex and each directly connected traceability vertex. In the supply chain relationship network, the attribute representation vector of the first traceability vertex is obtained by integrating the traceability vertex attribute feature vectors of its directly connected traceability vertices under a certain category in the previous steps, which reflects the comprehensive influence of the directly connected traceability vertices under this category on the first traceability vertex; while the traceability vertex attribute feature vector of the directly connected traceability vertex contains the basic information of this directly connected traceability vertex itself.

[0081] In step S332, based on the focusing influence coefficient of the first traceability vertex under any category, the attribute concatenation vector between the first traceability vertex and each directly connected traceability vertex is weighted and adjusted to obtain the adjusted attribute concatenation vector. The focusing influence coefficient of the first traceability vertex under a certain category reflects the importance of this category to the first traceability vertex. This coefficient is used to weight the attribute concatenation vector to highlight the characteristics of those directly connected traceability vertices that have a greater influence on the first traceability vertex.

[0082] In step S333, perform feature transformation on the adjusted attribute concatenation vector to obtain the focusing influence coefficients between the first traceability vertex and each directly connected traceability vertex. The purpose of feature transformation is to map the adjusted attribute concatenation vector to a new feature space, excavate the potential features and correlation information therein, so as to obtain the focusing influence coefficients that can better reflect the relationship between the first traceability vertex and the directly connected traceability vertices.

[0083] A linear transformation and a non - linear activation function can be used to achieve feature transformation. Assume that the dimension of the adjusted attribute concatenation vector is p, use a weight matrix W (with dimension q×p, where q is a pre - set dimension) and a bias vector b0 (with dimension q). First, perform a linear transformation to obtain the intermediate result e i , and the formula is . Then, use a non - linear activation function (such as the ReLU function, (x)=max(0, x)) to process e i to obtain the focusing influence coefficient f between the first traceability vertex and the i - th directly connected traceability vertex.i , the formula is .

[0084] In step S334, a normalization operation is performed on the focusing influence coefficient between the first traceability vertex and each directly connected traceability vertex to obtain the vertex dimension focusing influence coefficient between the first traceability vertex and each directly connected traceability vertex. The purpose of the normalization operation is to convert the focusing influence coefficient into a comparable and interpretable value. Usually, it is converted into a probability distribution so that the sum of all focusing influence coefficients is 1. A feasible normalization method is to use the Softmax function.

[0085] Through steps S331 - S334, the vertex dimension focusing influence coefficients between the first traceability vertex and each directly connected traceability vertex can be accurately obtained. These coefficients provide more accurate weight information for subsequent vertex feature extraction and anti-counterfeiting traceability detection, which helps to improve the performance and reliability of the entire anti-counterfeiting traceability system.

[0086] As an implementation, in step S400, based on the vertex dimension focusing influence coefficient of each traceability vertex, vertex feature extraction is performed on the traceability vertices in the supply chain relationship network to obtain the vertex coding vector of each traceability vertex, including: Step S410: Determine one or more directly connected traceability vertices of each traceability vertex, and determine the vertex dimension focusing influence coefficient between each traceability vertex and each neighbor; Step S420: Based on the vertex dimension focusing influence coefficient between each traceability vertex and each directly connected traceability vertex, perform feature transformation on the traceability vertex attribute feature vector of each directly connected traceability vertex to obtain the vertex coding vector of each traceability vertex.

[0087] In step S410, determine one or more directly connected traceability vertices of each traceability vertex, and determine the vertex dimension focusing influence coefficient between each traceability vertex and each neighbor (i.e., directly connected traceability vertex). In the supply chain relationship network, traceability vertices do not exist in isolation but are connected to other traceability vertices through various transfer relationships. These directly connected traceability vertices are the directly connected traceability vertices. The vertex dimension focusing influence coefficient reflects the importance of each directly connected traceability vertex to the target traceability vertex, which has been obtained through a series of dynamic weight assignment operations in the previous steps.

[0088] In step S420, based on the vertex dimension focusing influence coefficients between each traceability vertex and each directly connected traceability vertex, perform feature transformation on the traceability vertex attribute feature vectors of each directly connected traceability vertex to obtain the vertex encoding vector of each traceability vertex. The traceability vertex attribute feature vectors of directly connected traceability vertices contain the basic information of these directly connected traceability vertices, such as the production capacity of suppliers, product quality indicators, the transportation efficiency of logistics enterprises, etc. By combining the vertex dimension focusing influence coefficients to perform feature transformation on these attribute feature vectors, it is possible to highlight the characteristics of directly connected traceability vertices that have a greater impact on the target traceability vertex, thereby constructing a more representative vertex encoding vector.

[0089] A feasible feature transformation method is weighted summation. Assume that the target traceability vertex v has n directly connected traceability vertices , and the traceability vertex attribute feature vector of the i-th directly connected traceability vertex v i is x i , and the vertex dimension focusing influence coefficient between the target traceability vertex v and the i-th directly connected traceability vertex v i is , then the vertex encoding vector h v of the target traceability vertex v can be calculated by the following formula: .

[0090] In addition to weighted summation, a neural network-based method can also be used. A multi-layer perceptron (MLP) or graph neural network (GNN) model can be constructed, taking the traceability vertex attribute feature vectors of directly connected traceability vertices and the vertex dimension focusing influence coefficients as inputs, and through the non-linear transformation and feature extraction capabilities of the neural network, obtain a richer and more abstract vertex encoding vector.

[0091] In the actual implementation process, a deep learning framework (such as TensorFlow or PyTorch) can be used to implement the above feature transformation operations. For the weighted summation method, array operation functions can be used to implement the weighted sum and accumulation operations of vectors. For the neural network-based method, define the structure of the neural network, including the number of neurons and connection methods in the input layer, hidden layer, and output layer, and then use the training data to train the model. Adjust the parameters of the model through the backpropagation algorithm so that the model can accurately extract meaningful features for the target traceability vertex from the attribute feature vectors of directly connected traceability vertices, and finally obtain the vertex encoding vector.

[0092] In addition, post-processing is also performed on the obtained vertex encoding vectors, such as normalization operations, to ensure the comparability and stability of the vertex encoding vectors of different traceability vertices. Feasible normalization methods include L1 normalization and L2 normalization. Taking L2 normalization as an example, for the vertex encoding vector , the normalized vertex encoding vector It can be calculated by the following formula: wherein, is the L2 norm of the vertex encoding vector h v , and the calculation formula is is the dimension of the vertex encoding vector.

[0093] Through steps S410 and S420, the direct-connected traceable vertices of the traceable vertex and their vertex dimension focusing influence coefficients can be accurately determined, and based on this information, the attribute feature vectors of the direct-connected traceable vertices are subjected to feature transformation to obtain representative vertex encoding vectors. These vertex encoding vectors will provide key feature information for subsequent anti-counterfeiting traceability detection, help improve the accuracy and reliability of anti-counterfeiting traceability, and ensure the safe and stable operation of the supply chain.

[0094] As an implementation manner, for the foregoing step S500, based on the vertex encoding vector of each traceable vertex, anti-counterfeiting traceability detection is performed on each traceable vertex to obtain the anti-counterfeiting traceability detection result of each traceable vertex, and it can also be implemented in the following manner: Step S510: Combine the vertex encoding vector of each traceable vertex and the traceable vertex attribute feature vector of each traceable vertex to obtain the combined vector of each traceable vertex; Step S520: Perform a fully connected mapping based on the forged traceable vertex on the combined vector of each traceable vertex to obtain the forgery probability of each traceable vertex; Step S530: Determine the anti-counterfeiting traceability detection result of each traceable vertex based on the forgery probability of each traceable vertex.

[0095] In step S510, the vertex encoding vector of each traceable vertex and the traceable vertex attribute feature vector of each traceable vertex are combined (such as splicing) to obtain the combined vector of each traceable vertex. The vertex encoding vector is obtained through the previous steps by performing feature transformation on the attribute feature vector of the direct-connected traceable vertex based on the vertex dimension focusing influence coefficient, and it contains the comprehensive feature information of the traceable vertex in the entire supply chain relationship network; while the traceable vertex attribute feature vector is the feature expression of the basic information of the traceable vertex and reflects the inherent attributes of the vertex itself. Combining these two vectors can make full use of their respective information and provide a more comprehensive and richer feature basis for subsequent anti-counterfeiting traceability detection.

[0096] In step S520, a fully connected mapping based on forged traceability vertices is performed on the combined vector of each traceability vertex to obtain the forgery probability of each traceability vertex. The fully connected mapping is a feasible neural network operation that maps the input combined vector to a new feature space through a series of linear transformations and non-linear activation functions, thereby obtaining the probability that the traceability vertex is forged. In this process, a pre-trained neural network model is used. The input of this model is the combined vector of the traceability vertex, and the output is a probability value between 0 and 1, indicating the possibility that the traceability vertex is forged.

[0097] The fully connected neural network model contains multiple fully connected layers. Each fully connected layer consists of multiple neurons, and the neurons are connected by weights. Suppose the combined vector c is input into the first fully connected layer, which has p neurons. The output of each neuron is the weighted sum of the input vector and the corresponding weight plus a bias term, and then it is processed by a non-linear activation function (such as the ReLU function, ReLU(x) = max(0, x)). The output of the first fully connected layer is then used as the input of the next fully connected layer, and so on, until the last fully connected layer. The last fully connected layer usually uses the Sigmoid function as the activation function to map the output value to the interval [0, 1] to obtain the forgery probability P of the lipstick traceability vertex. Suppose the weight matrix of the i-th fully connected layer is W i , the bias vector is b i , the input vector is x i , then the output x i+1 of this layer can be calculated by the following formula: , where, is the activation function. For the last fully connected layer, the Sigmoid function is used to obtain the forgery probability P.

[0098] In step S530, based on the forgery probability of each traceability vertex, the anti-counterfeiting traceability detection result of each traceability vertex is determined. A pre-set threshold is used to judge whether the traceability vertex is forged. If the forgery probability of a certain traceability vertex is greater than the threshold, it is determined that the traceability vertex is forged; otherwise, it is determined to be genuine.

[0099] As an implementation manner, the method provided by this application is implemented based on multiple sequentially connected weight assignment focusing components. The foregoing steps S200~S400 can be specifically implemented by the following process: The following operations are performed based on the x-th weight assignment focusing component: Step S10: Based on the vertex encoding vectors of the traceability vertices loaded into the x-th weight assignment focusing component, perform a dynamic weight assignment operation on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices, to obtain the type dimension focusing influence coefficients of each traceability vertex in the x-th weight assignment focusing component, where x is a positive integer, the vertex encoding vector of the traceability vertex loaded into the first weight assignment focusing component is the traceability vertex attribute feature vector of the traceability vertex, and the vertex encoding vector of the traceability vertex loaded into other weight assignment focusing components is the vertex encoding vector of the traceability vertex output by the previous weight assignment focusing component of other weight assignment focusing components; Step S20: Based on the type dimension focusing influence coefficients of each traceability vertex in the x-th weight assignment focusing component, perform a vertex dimension dynamic weight assignment operation on each traceability vertex, to obtain the vertex dimension focusing influence coefficients of each traceability vertex in the x-th weight assignment focusing component; Step S30: Based on the vertex dimension focusing influence coefficients of each traceability vertex in the x-th weight assignment focusing component, perform vertex feature extraction on the traceability vertices in the supply chain relationship network, to obtain the vertex encoding vectors of each traceability vertex in the x-th weight assignment focusing component; Step S40: Based on the vertex encoding vectors of each traceability vertex in the last weight assignment focusing component, perform anti-counterfeiting traceability detection on each traceability vertex, to obtain the anti-counterfeiting traceability detection results of each traceability vertex.

[0100] In Step S10, based on the vertex encoding vectors of the traceability vertices loaded into the x-th weight assignment focusing component, perform a dynamic weight assignment operation on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices, to obtain the type dimension focusing influence coefficients of each traceability vertex in the x-th weight assignment focusing component. Here, x is a positive integer, the vertex encoding vector of the traceability vertex loaded into the first weight assignment focusing component is the traceability vertex attribute feature vector of the traceability vertex, and the vertex encoding vector of the traceability vertex loaded into other weight assignment focusing components is the vertex encoding vector of the traceability vertex output by the previous weight assignment focusing component of other weight assignment focusing components. The attention mechanism can be used to achieve this dynamic weight assignment. Assume that the target traceability vertex is v, and its i-th directly connected traceability vertex is v i , the attention score e i can be obtained by calculating the similarity between the query vector q of the target traceability vertex and the key vector k i of the directly connected traceability vertex. For example, , feasible similarity calculation methods include dot product, cosine similarity, etc. Then, the attention score is converted into a weight through the Softmax function , where n is the number of directly connected traceable vertices. By performing such weight assignments for different types of directly connected traceable vertices, the focus influence coefficients of the vehicle traceable vertex under each type can be obtained, and the set of these coefficients is the focus influence coefficient in the type dimension.

[0101] In step S20, based on the focus influence coefficients in the type dimension of each traceable vertex in the x-th weight assignment focus component, a vertex dimension dynamic weight assignment operation is performed on each traceable vertex to obtain the focus influence coefficients in the vertex dimension of each traceable vertex in the x-th weight assignment focus component. After obtaining the focus influence coefficients in the type dimension, the influence of each specific directly connected traceable vertex on the target traceable vertex is further considered.

[0102] In step S30, based on the focus influence coefficients in the vertex dimension of each traceable vertex in the x-th weight assignment focus component, vertex feature extraction is performed on the traceable vertices in the supply chain relationship network to obtain the vertex coding vectors of each traceable vertex in the x-th weight assignment focus component. After obtaining the focus influence coefficients in the vertex dimension, these coefficients are used to perform weighted combination on the attribute feature vectors of the directly connected traceable vertices, so as to extract more characteristic information that can represent the target traceable vertex.

[0103] In step S40, based on the vertex coding vectors of each traceable vertex in the last weight assignment focus component, anti-counterfeiting traceability detection is performed on each traceable vertex to obtain the anti-counterfeiting traceability detection results of each traceable vertex. After being processed by multiple weight assignment focus components, the vertex coding vectors output by the last component have fully integrated the information of each link in the supply chain relationship network, and these vectors are used for anti-counterfeiting traceability detection.

[0104] A classification model, such as a neural network classifier, can be used, taking the vertex coding vector output by the last weight assignment focus component as the input. Assuming the vertex coding vector is h, the input layer of the neural network classifier receives h, undergoes feature transformation and extraction through the intermediate hidden layer, and finally obtains the probability P that the traceable vertex is forged through the activation function (such as the Sigmoid function) of the output layer. For example, the output of the neural network , where W1 and W2 are weight matrices, b1 and b2 are bias vectors, ReLU is the activation function, is the Sigmoid function. According to a preset threshold (such as 0.5) to determine whether the traceable vertex is forged. If P > 0.5, it is determined to be forged; otherwise, it is determined to be genuine.

[0105] In the actual implementation process, deep learning frameworks (such as TensorFlow, PyTorch) can be used to implement the above steps. For dynamic weight allocation, the attention mechanism module provided in the framework can be used; for vertex feature extraction, weighted summation can be achieved through custom layers and tensor operations; for anti-counterfeiting traceability detection, a neural network classifier can be constructed and trained. It is also possible to jointly train the parameters of multiple weight allocation focusing components, and adjust the weights and biases in each component through the backpropagation algorithm to improve the performance of the entire anti-counterfeiting traceability system. Through steps S10 - S40, the information of the traceability vertices in the supply chain relationship network can be gradually and deeply explored. After multiple rounds of weight allocation and feature extraction, anti-counterfeiting traceability detection can be accurately performed finally to ensure the security of the supply chain and the authenticity of the products.

[0106] As an implementation manner, the vertex encoding vector of each traceability vertex is obtained by processing based on a dynamic weight allocation strategy. The method provided in this application further includes: Perform the following operations on the first traceability vertex in the supply chain relationship network: Step S1: Integrate the vertex encoding vectors of each traceability vertex obtained based on multiple dynamic weight allocation strategies respectively to obtain the integrated vertex encoding vector of each traceability vertex.

[0107] Based on this, in the above step S500, based on the vertex encoding vector of each traceability vertex, anti-counterfeiting traceability detection is performed on each traceability vertex to obtain the anti-counterfeiting traceability detection result of each traceability vertex, including: Step S501: Based on the integrated vertex encoding vector of each traceability vertex, anti-counterfeiting traceability detection is performed on each traceability vertex to obtain the anti-counterfeiting traceability detection result of each traceability vertex.

[0108] In step S1, the vertex encoding vectors of each traceability vertex obtained based on multiple dynamic weight allocation strategies respectively are integrated to obtain the integrated vertex encoding vector of each traceability vertex. In the entire anti-counterfeiting traceability process, multiple different dynamic weight allocation strategies may be adopted to generate vertex encoding vectors. Each strategy explores and processes the information of the traceability vertex and its directly connected traceability vertices from different perspectives. However, the vertex encoding vectors generated by a single strategy may have problems such as incomplete or inaccurate information. Therefore, the vertex encoding vectors obtained based on multiple strategies are integrated to obtain richer and more accurate feature information. A feasible method for integration operation is weighted summation. Another integration method is concatenation, where the vertex encoding vectors obtained based on different strategies are concatenated in sequence to form a longer vector, which is not specifically limited.

[0109] Based on the integrated vertex coding vectors obtained in step S1, in step S501, for each traceable vertex, anti-counterfeiting traceability detection is performed based on the integrated vertex coding vector of each traceable vertex, and the anti-counterfeiting traceability detection result of each traceable vertex is obtained. The integrated vertex coding vector synthesizes the information of various dynamic weight allocation strategies and more comprehensively reflects the characteristics of the traceable vertex in the supply chain relationship network. Therefore, the accuracy of anti-counterfeiting traceability detection can be improved.

[0110] In the embodiments of the present application, a classification model can be used for anti-counterfeiting traceability detection. Taking a neural network classifier as an example, the integrated vertex coding vector is used as the input, and feature transformation and extraction are performed through multiple layers of the neural network. Assume that the integrated vertex coding vector is , the input layer of the neural network receives , the intermediate hidden layer performs a non-linear transformation on the input to learn a higher-level feature representation. Finally, the output layer outputs the probability P that the traceable vertex is forged through an activation function (such as the Sigmoid function). Assume that the structure of the neural network is , where W1 and W2 are weight matrices, b1 and b2 are bias vectors, ReLU is the activation function, is the Sigmoid function. According to a preset threshold (such as 0.5) to determine whether the traceable vertex is forged. If P > 0.5, it is determined that the traceable vertex is forged; otherwise, it is determined to be genuine. For example, for the above lipstick traceable vertex, if the forged probability P = 0.7 obtained through the neural network classifier, since 0.7 > 0.5, it is determined that the lipstick traceable vertex is forged; if P = 0.3, it is determined to be genuine.

[0111] Through step S1 and step S501, the information of various dynamic weight allocation strategies can be fully utilized to obtain more accurate integrated vertex coding vectors, and anti-counterfeiting traceability detection is performed based on this, thereby improving the performance and reliability of the entire anti-counterfeiting traceability system.

[0112] The embodiments of the present application provide a computer system, as Figure 2 shown, the computer system 100 includes: a processor 101 and a memory 103. Among them, the processor 101 and the memory 103 are connected, such as connected through a bus 102. Optionally, the computer system 100 may further include a transceiver 104. It should be noted that in actual applications, the transceiver 104 is not limited to one, and the structure of the computer system 100 does not constitute a limitation to the embodiments of the present application.

[0113] The processor 101 can be a CPU, a general-purpose processor, a GPU, a DSP, an ASIC, an FPGA, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logical blocks, modules, and circuits described in connection with the disclosure of this application. The processor 101 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0114] The bus 102 can include a path for transmitting information between the above components. The bus 102 can be a PCI bus, an EISA bus, or the like. The bus 102 can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 2 only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0115] The memory 103 can be a ROM or other type of static storage device that can store static information and instructions, a RAM, or other type of dynamic storage device that can store information and instructions. It can also be an EEPROM, a CD-ROM, or other optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk storage medium, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0116] The memory 103 is used to store the application program code for executing the solution of this application, and is controlled by the processor 101 to execute. The processor 101 is used to execute the application program code stored in the memory 103 to implement the content shown in any of the foregoing method embodiments.

[0117] The embodiments of this application provide a computer system. The computer system in the embodiments of this application includes: one or more processors; a memory; one or more computer programs, where one or more computer programs are stored in the memory and are configured to be executed by one or more processors. When the one or more programs are executed by the processor, the methods provided in the above embodiments are implemented.

[0118] The embodiments of this application provide a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program runs on the processor, the processor can execute the corresponding content in the foregoing method embodiments.

[0119] It should be understood that although the steps in the flowchart of the accompanying drawings are shown sequentially according to the indication of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear indication in this document, there is no strict order restriction for the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

[0120] The above are only some embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. An anti-counterfeiting and tracing method based on big data analysis, characterized in that: The method comprises: Acquire a supply chain relationship network including a plurality of traceability vertices, wherein the type of the traceability vertices is a product traceability vertex or a flow traceability vertex; Performing a dynamic weight allocation operation on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices to obtain a type dimension focusing influence coefficient of each traceability vertex; Based on the category dimension focusing influence coefficient of each of the traceability vertices, a vertex dimension dynamic weight allocation operation is performed on each of the traceability vertices to obtain the vertex dimension focusing influence coefficient of each of the traceability vertices; Based on the vertex dimension focusing influence coefficient of each traceability vertex, vertex feature extraction is performed on the traceability vertex in the supply chain relationship network to obtain a vertex encoding vector of each traceability vertex; Based on the vertex encoding vector of each of the traceability vertices, an anti-counterfeiting traceability detection is performed on each of the traceability vertices to obtain an anti-counterfeiting traceability detection result of each of the traceability vertices.

2. The method according to claim 1, characterized in that Before performing a dynamic weight allocation operation on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices to obtain the type dimension focusing influence coefficient of each traceability vertex, the method further includes: Based on one or more flow situations between the product traceability vertex and the flow traceability vertex, an influence coefficient of a connection line between the product traceability vertex and the flow traceability vertex is determined.

3. The method according to claim 2, characterized in that The determining, based on one or more flow situations between the product traceability vertex and the flow traceability vertex, an influence coefficient of a connection line between the product traceability vertex and the flow traceability vertex includes: Obtaining influence coefficients corresponding to one or more flow situations between the product traceability vertex and the flow traceability vertex; The influence coefficients corresponding to one or more flow situations between the product traceability vertex and the flow traceability vertex are fused to obtain the influence coefficient of the connecting line between the product traceability vertex and the flow traceability vertex.

4. The method according to claim 1, characterized in that: The step of performing a dynamic weight allocation operation on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices to obtain a type dimension focus influence coefficient of each traceability vertex includes: The following operations are performed on the first traceability vertex in the supply chain relationship network: Determine one or more directly connected traceability vertices of any type of the first traceability vertex, wherein the one or more directly connected traceability vertices of any type are traceability vertices connected to the first traceability vertex based on the flow situation of any type; Performing an integration operation on the traceability vertex attribute feature vectors of the one or more directly connected traceability vertices to obtain an attribute representation vector of the first traceability vertex under any of the categories; Performing a dynamic weight allocation operation on the attribute representation vector of the first tracing vertex under any of the categories and the tracing vertex attribute feature vector of the first tracing vertex to obtain a focusing influence coefficient of the first tracing vertex under any of the categories; The set of focusing influence coefficients of the first tracing vertex under the multiple categories is used as the category dimension focusing influence coefficient of each tracing vertex.

5. The method according to claim 4, characterized in that The integrating operation of the attribute feature vectors of the one or more directly connected traceable vertices to obtain the attribute representation vector of the first traceable vertex under any category includes: Determine an influence coefficient of a connection line between each of the directly connected traceability vertices and the first traceability vertex; Based on the influence coefficient of the connection line between each of the directly connected traceable vertices and the first traceable vertex, a fusion operation is performed on the traceable vertex attribute feature vector of each of the directly connected traceable vertices to obtain an attribute representation vector of the first traceable vertex under any of the categories; The performing a dynamic weight allocation operation on the attribute representation vector of the first tracing vertex under any category and the tracing vertex attribute feature vector of the first tracing vertex to obtain a focusing influence coefficient of the first tracing vertex under any category includes: splicing the attribute representation vector of the first tracing vertex under the any category and the tracing vertex attribute feature vector of the first tracing vertex to obtain an attribute splicing vector of the first tracing vertex under the any category; Based on the adaptive variables of any one category, feature transformation is performed on the attribute concatenation vector of the first traceability vertex under any one category to obtain a focusing influence coefficient of the first traceability vertex under any one category; A focusing influence coefficient of the first tracing vertex under any one of the categories is normalized to obtain the focusing influence coefficient of the first tracing vertex under any one of the categories.

6. The method according to claim 1, characterized in that The step of performing a vertex dimension dynamic weight allocation operation on each of the tracing vertices based on the category dimension focusing influence coefficient of each of the tracing vertices to obtain the vertex dimension focusing influence coefficient of each of the tracing vertices includes: The following operations are performed on the first traceability vertex in the supply chain relationship network: Determine one or more directly connected traceability vertices of any type of the first traceability vertex, wherein the one or more directly connected traceability vertices of any type are traceability vertices connected to the first traceability vertex based on the flow situation of any type; Obtaining a focusing influence coefficient of the first tracing vertex in any one of the categories from the category-dimensional focusing influence coefficient of the first tracing vertex; Based on the focusing influence coefficient of the first tracing vertex under any of the categories and the one or more directly connected tracing vertices, a dynamic weight allocation operation is performed on the first tracing vertex to obtain the vertex dimension focusing influence coefficient of the first tracing vertex.

7. The method according to claim 6, characterized in that The step of performing a dynamic weight allocation operation on the first tracing vertex based on the focusing influence coefficient of the first tracing vertex under any of the categories and the one or more directly connected tracing vertices to obtain the vertex dimension focusing influence coefficient of the first tracing vertex includes: Splicing the attribute representation vector of the first traceability vertex and the traceability vertex attribute feature vector of each directly connected traceability vertex to obtain an attribute splicing vector between the first traceability vertex and each directly connected traceability vertex; Based on the focusing influence coefficient of the first tracing vertex under any of the categories, weighted adjustment is performed on the attribute splicing vector between the first tracing vertex and each of the directly connected tracing vertices to obtain the adjusted attribute splicing vector; Performing feature transformation on the adjusted attribute splicing vector to obtain a focusing influence coefficient between the first traceability vertex and each of the directly connected traceability vertices; A normalization operation is performed on the focusing influence coefficient between the first tracing vertex and each of the directly connected tracing vertices to obtain a vertex dimension focusing influence coefficient between the first tracing vertex and each of the directly connected tracing vertices.

8. The method according to claim 1, characterized in that The step of extracting vertex features of the traceability vertices in the supply chain relationship network based on the vertex dimension focusing influence coefficient of each traceability vertex to obtain a vertex encoding vector of each traceability vertex includes: Determine one or more directly connected traceable vertices of each of the traceable vertices, and determine a vertex dimension focusing influence coefficient between each of the traceable vertices and each of the neighbors; Based on the vertex dimension focusing influence coefficient between each of the traceable vertices and each of the directly connected traceable vertices, feature transformation is performed on the traceable vertex attribute feature vector of each of the directly connected traceable vertices to obtain a vertex encoding vector of each of the traceable vertices; The anti-counterfeiting and tracing detection is performed on each of the tracing vertices based on the vertex encoding vector of each of the tracing vertices to obtain the anti-counterfeiting and tracing detection result of each of the tracing vertices, including: Combining the vertex encoding vector of each of the tracing-back vertex and the tracing-back vertex attribute feature vector of each of the tracing-back vertex to obtain a combined vector of each of the tracing-back vertex; Performing a full-connection mapping based on the forged tracing vertex on the combined vector of each of the tracing vertices to obtain the forgery probability of each of the tracing vertices; Based on the forgery probability of each of the traceability vertices, determining the anti-counterfeiting traceability detection result of each of the traceability vertices; The method is implemented based on a plurality of weight allocation focusing components connected in sequence, wherein a dynamic weight allocation operation is performed on the traceability vertices in the supply chain relationship network based on the types of directly connected traceability vertices to obtain a type dimension focusing influence coefficient of each traceability vertex; based on the type dimension focusing influence coefficient of each traceability vertex, a vertex dimension dynamic weight allocation operation is performed on each traceability vertex to obtain a vertex dimension focusing influence coefficient of each traceability vertex; based on the vertex dimension focusing influence coefficient of each traceability vertex, vertex feature extraction is performed on the traceability vertices in the supply chain relationship network to obtain a vertex encoding vector of each traceability vertex; based on the vertex encoding vector of each traceability vertex, an anti-counterfeiting traceability detection is performed on each traceability vertex to obtain an anti-counterfeiting traceability detection result of each traceability vertex, including: Based on the xth weight allocation focus component performs the following operations: Based on the vertex encoding vector of the traceability vertex loaded into the x-th weight allocation focusing component, a dynamic weight allocation operation based on the types of directly connected traceability vertices is performed on the traceability vertices in the supply chain relationship network to obtain the type dimension focusing influence coefficient of each traceability vertex in the x-th weight allocation focusing component, where x is a positive integer, the vertex encoding vector of the traceability vertex loaded into the first weight allocation focusing component is the traceability vertex attribute feature vector of the traceability vertex, and the vertex encoding vector of the traceability vertex loaded into other weight allocation focusing components is the vertex encoding vector of the traceability vertex output by the previous weight allocation focusing component of other weight allocation focusing components; Based on the category dimension focusing influence coefficient of each traceability vertex in the xth weight allocation focusing component, a vertex dimension dynamic weight allocation operation is performed on each traceability vertex to obtain the vertex dimension focusing influence coefficient of each traceability vertex in the xth weight allocation focusing component; Based on the vertex dimension focusing influence coefficient of each traceability vertex in the xth weight allocation focusing component, vertex feature extraction is performed on the traceability vertex in the supply chain relationship network to obtain a vertex encoding vector of each traceability vertex in the xth weight allocation focusing component; Based on the vertex encoding vector of each traceability vertex in the last weight distribution focusing component, anti-counterfeiting traceability detection is performed on each traceability vertex to obtain the anti-counterfeiting traceability detection result of each traceability vertex.

9. The method according to claim 1, characterized in that: The vertex encoding vector of each of the traceable vertices is obtained based on a dynamic weight allocation strategy; The method further comprises: The following operations are performed on the first traceability vertex in the supply chain relationship network: Performing an integration operation on the vertex encoding vectors of each of the source tracing vertices respectively obtained based on a plurality of the dynamic weight allocation strategies to obtain an integrated vertex encoding vector of each of the source tracing vertices; The anti-counterfeiting and tracing detection is performed on each of the tracing vertices based on the vertex encoding vector of each of the tracing vertices to obtain the anti-counterfeiting and tracing detection result of each of the tracing vertices, including: Based on the integrated vertex encoding vector of each of the traceability vertices, an anti-counterfeiting traceability detection is performed on each of the traceability vertices to obtain an anti-counterfeiting traceability detection result of each of the traceability vertices.

10. A computer system, characterized in that: include: one or more processors; Memory; one or more computer programs; The one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method according to any one of claims 1 to 9 is implemented.

Images