Supply chain data sharing method for cross-enterprise collaboration

Abstract

The present application relates to a supply chain data sharing method for cross-enterprise collaboration, aiming at the problems of lagging static permission management and privacy leakage in verification process, the scheme constructs a dynamic trust graph, fuses multi-source heterogeneous data to generate an initial trust score; using zero-knowledge proof protocol, the environmental compliance verification is completed without leaking the geographic location and device fingerprint; real-time monitoring of business flow events drives the dynamic evolution of trust score, and the data desensitization granularity is adaptively adjusted accordingly; combined with the smart contract of the block chain, the automatic freezing of violation is realized. The present application realizes the real-time response and fine-grained control of permission management, and effectively balances the data sharing efficiency and privacy security.

Description

Technical Field

[0001] This invention relates to the field of cross-organizational data security collaboration technology based on dynamic trust graphs and zero-knowledge proofs, and in particular to a supply chain data sharing method for cross-enterprise collaboration. Background Technology

[0002] As global supply chain networks become more complex and digitalized, cross-enterprise collaboration has become crucial for improving logistics efficiency and response speed. Data sharing, as a core element of collaboration, is increasingly valued for its security and privacy protection.

[0003] Current mainstream technical solutions in this field primarily rely on role-based access control models or static contractual constraint mechanisms. These mechanisms use pre-defined permission lists to restrict the scope of access to supply chain data by different enterprises. Some advanced systems are beginning to introduce blockchain technology to achieve traceability of data flow, utilizing distributed ledgers to record data exchange logs to prevent tampering. In scenarios involving the sharing of sensitive business information such as inventory levels, production plans, and logistics trajectories, industry practice typically employs full encryption transmission or simple field masking to attempt to find a balance between data availability and confidentiality.

[0004] However, in existing typical technology application scenarios, trust assessment is often a one-off or static qualification-based process, lacking the dynamic perception capability of partners' real-time performance, order status changes, and abnormal events. This results in trust scores not being able to adjust in real time with changes in business flow. Static trust mechanisms are ill-equipped to cope with the rapidly changing risks in the supply chain environment. Once initial authorization is completed, even if the partner shows signs of default or the operating environment is abnormal, the system may still continue to keep highly sensitive data interfaces open, posing a risk of data leakage. Secondly, existing access control policies are coarse-grained, often adopting a "black and white" data delivery model, either completely opening plaintext or completely denying access. They lack the fine-grained control capability to dynamically adjust the data anonymization strength and aggregation granularity according to the real-time trust level, resulting in over-exposure risks in low-trust scenarios or insufficient data value mining in high-trust scenarios. Summary of the Invention

[0005] This application provides a supply chain data sharing method for cross-enterprise collaboration, aiming to solve one of the problems or issues of the existing technology mentioned in the background section.

[0006] This application provides a supply chain data sharing method for cross-enterprise collaboration, specifically including:

[0007] S1: Obtain enterprise qualification data, historical performance records and third-party credit scores of each participant in the cross-enterprise collaboration scenario, generate an initial trust score, and use the initial trust score as the initial attribute of the dynamic trust graph node to construct the dynamic trust graph.

[0008] S2: Receives access requests initiated by data requesters, generates zero-knowledge proof credentials based on local sensitive context information using zero-knowledge proof protocols, and verifies the compliance of the operating environment without disclosing specific geographical locations and device fingerprints.

[0009] S3: Monitor order status changes, logistics node updates, and abnormal alarm event signals in the supply chain business flow in real time, input the event signals as weighting factors into the dynamic trust graph, calculate and output the real-time updated dynamic trust score.

[0010] S4: Based on the numerical range of the dynamic trust score, match the corresponding data anonymization algorithm strength level and field visibility range strategy to generate a fine-grained access control parameter group containing masking rules or aggregation granularity.

[0011] S5: Based on the fine-grained access control parameter group, perform differentiated processing on the target supply chain data. If the dynamic trust score is higher than the preset high trust threshold, output complete plaintext data. If it is lower than the preset low trust threshold, output trend data that has been masked or aggregated by key values.

[0012] S6: Perform hash operations on the access request details, zero-knowledge proof verification results, dynamic trust score change trajectory, and final data delivery form to generate tamper-proof audit evidence data and write it into the blockchain ledger.

[0013] S7: Monitor unauthorized access patterns in audit and evidence storage data. If potential risky behavior is detected, trigger the smart contract to execute a trust score freeze instruction.

[0014] This application provides a supply chain data sharing method for cross-enterprise collaboration, which has the following beneficial effects:

[0015] By constructing a multi-dimensional dynamic trust graph and a real-time evolving permission management mechanism, continuous and fine-grained control of access permissions for cross-enterprise supply chain participants can be achieved, effectively avoiding over-authorization or permission lag caused by fixed permission configurations, and significantly improving the flexibility and security of data access control in complex and ever-changing collaborative business scenarios.

[0016] By introducing a context verification mechanism based on zero-knowledge proof, the compliance verification of the operating environment can be completed without exposing sensitive details such as the requester's location information and device fingerprint. This overcomes the technical contradiction between privacy protection and permission determination in existing solutions, significantly improves the efficiency of trust establishment in multi-party collaboration, and reduces the risk of sensitive information leakage.

[0017] By combining smart contract-driven event response mechanisms with blockchain's evidence storage and traceability capabilities, dynamic events in the supply chain business flow are used as trust weight factors to update node scores in real time. All access behaviors and decision trajectories are recorded in an immutable manner. This not only enables the permission system to adapt and self-heal from risks, but also provides a complete and reliable data chain for regulatory audits, significantly enhancing the system's traceability and anti-attack capabilities. Attached Figure Description

[0018] Figure 1 This is the main flowchart of a supply chain data sharing method for cross-enterprise collaboration.

[0019] Figure 2 This is a sub-flowchart of a supply chain data sharing method for cross-enterprise collaboration.

[0020] Figure 3 This is another sub-flowchart of a supply chain data sharing approach oriented towards cross-enterprise collaboration.

[0021] Figure 4 This is an application environment diagram of a supply chain data sharing method for cross-enterprise collaboration in one embodiment of this application.

[0022] Figure 5 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application. Detailed Implementation

[0023] The following disclosed embodiments or examples are used to implement the present invention. Of course, they are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0024] like Figure 1 As shown, this application provides a supply chain data sharing method for cross-enterprise collaboration, specifically including:

[0025] S1: Obtain enterprise qualification data, historical performance records and third-party credit scores of each participant in the cross-enterprise collaboration scenario, generate an initial trust score, and use the initial trust score as the initial attribute of the dynamic trust graph node to construct the dynamic trust graph.

[0026] S2: Receives access requests initiated by data requesters, generates zero-knowledge proof credentials based on local sensitive context information using zero-knowledge proof protocols, and verifies the compliance of the operating environment without disclosing specific geographical locations and device fingerprints.

[0027] S3: Monitor order status changes, logistics node updates, and abnormal alarm event signals in the supply chain business flow in real time, input the event signals as weighting factors into the dynamic trust graph, calculate and output the real-time updated dynamic trust score.

[0028] S4: Based on the numerical range of the dynamic trust score, match the corresponding data anonymization algorithm strength level and field visibility range strategy to generate a fine-grained access control parameter group containing masking rules or aggregation granularity.

[0029] S5: Based on the fine-grained access control parameter group, perform differentiated processing on the target supply chain data. If the dynamic trust score is higher than the preset high trust threshold, output complete plaintext data. If it is lower than the preset low trust threshold, output trend data that has been masked or aggregated by key values.

[0030] S6: Perform hash operations on the access request details, zero-knowledge proof verification results, dynamic trust score change trajectory, and final data delivery form to generate tamper-proof audit evidence data and write it into the blockchain ledger.

[0031] S7: Monitor unauthorized access patterns in audit and evidence storage data. If potential risky behavior is detected, trigger the smart contract to execute a trust score freeze instruction.

[0032] Step S1: Obtain enterprise qualification data, historical performance records, and third-party credit scores of each participant in the cross-enterprise collaboration scenario, generate an initial trust score, and use the initial trust score as the initial attribute of the dynamic trust graph node to construct the dynamic trust graph. Specifically, this includes:

[0033] S1.1: Obtain enterprise qualification certificate texts, historical performance transaction logs, and third-party credit scoring reports from each participant in the cross-enterprise collaboration scenario. Use natural language processing technology to extract entities from the unstructured qualification texts and perform time series alignment processing on the transaction logs to generate a standardized set of multi-dimensional original feature vectors.

[0034] For unstructured enterprise qualification documents, discrete historical performance transaction logs, and heterogeneous third-party credit scoring reports provided by various participants in cross-enterprise collaboration scenarios, a multi-source heterogeneous data standardization process is initiated to generate a multi-dimensional original feature vector set. A pre-trained named entity recognition model is invoked to perform semantic parsing on the qualification certificate text, extracting key static attribute fields such as registered capital, business scope, and certification level, transforming the unstructured text into a structured sequence of static enterprise features. Historical performance transaction log data is read, and dynamic behavioral records such as transaction timestamps, order amounts, delivery delay durations, and default markers are parsed to construct a dynamic behavioral event stream of the enterprise arranged along a timeline.

[0035] A time series alignment algorithm combining linear interpolation and nearest neighbor filling is introduced (linear interpolation is used when the number of consecutive missing points is ≤3, and nearest neighbor filling is used otherwise). This algorithm maps the non-uniformly sampled transaction logs caused by differences in business frequency among different participants to a unified standard time granularity grid, thereby eliminating the distribution bias in the time dimension.

[0036] The format of third-party credit scoring reports is normalized, and the quantitative score values ​​provided by authoritative institutions are extracted and embedded into the feature space as external verification factors. The extracted static feature sequences of enterprises, the aligned dynamic behavioral event streams, and the external verification factors are concatenated into vectors to construct a high-dimensional original feature matrix containing static reputation, dynamic performance, and external credit dimensions. Through data type conversion and missing value cleaning mechanisms, it is ensured that each element in the matrix is ​​a computable numerical data, forming a standardized set of multidimensional original feature vectors. Through the above multi-source data parsing, time series alignment, and feature fusion processing methods, the chaotic and heterogeneous original data obtained in the previous step is transformed into a standardized set of multidimensional original feature vectors with unified structure and time alignment. This achieves semantic consistency and computational usability of cross-enterprise data in the feature space, providing high-quality input for subsequent entropy weight calculation.

[0037] S1.2: Based on the standardized multidimensional original feature vector set, the objective weight coefficients of each dimension feature are calculated using the entropy weight method, and weighted fusion calculation is performed in combination with the preset subjective weight parameters of industry experts to output a comprehensive credit evaluation factor that includes enterprise static credit indicators and dynamic behavior indicators.

[0038] Receive the standardized multidimensional raw feature vector set output from step S1.1. This set contains numerical data on the static reputation dimension, dynamic performance dimension, and external credit dimension of each participating enterprise, serving as the input for the entropy weight method calculation and subjective weight fusion. Construct a feature information entropy calculation model, traversing each dimension feature in the feature vector set, statistically analyzing the distribution probability of all enterprise sample data under that dimension, and using a logarithmic function to calculate the dispersion index of each dimension feature, generating a sequence of raw information entropy values ​​reflecting data variability.

[0039] Based on information theory principles, a smaller information entropy indicates a greater discriminative power provided by that dimension. A normalization coefficient is used to perform a reverse mapping process on the original information entropy value, calculating the objective weight coefficients for each dimension feature to ensure the total weight is 1 and entirely determined by the inherent distribution of the data. A pre-built industry expert knowledge base is invoked to retrieve subjective weight parameter vectors preset for different business scenarios in the supply chain. These vectors reflect prior judgments in industry experience regarding the importance of qualifications, performance history, and external credit. A subjective-objective weight fusion mechanism is established, introducing a balancing factor to adjust the contribution ratio of objective data-driven weights and subjective experience-driven weights. A linear weighting operation is performed to generate a comprehensive weight vector, preserving the objective differences in the data while incorporating the decision preferences of domain experts. The generated comprehensive weight vector is then multiplied by a matrix multiplication operation with the standardized multidimensional original feature vector set. The feature values ​​of each dimension are then weighted and aggregated to output a comprehensive credit assessment factor scalar representing the overall credit level of the enterprise. The information entropy value of the j-th dimension feature is calculated using the following formula:

[0040]

[0041] Among them, E j Let p be the information entropy of the j-th dimension feature, n be the total number of participating enterprise samples, and p be the information entropy of the j-th dimension feature. i Let be the normalized probability value of the i-th enterprise on the j-th dimension feature. The objective weight coefficients are derived based on the information entropy calculation results.

[0042]

[0043] Among them, w j Let be the objective weight of the j-th feature dimension, and m be the total number of feature dimensions. The denominator is the sum of the information utility values ​​of all dimensions to achieve normalization. Perform a subjective-objective weight fusion operation to generate a comprehensive weight:

[0044]

[0045] Among them, W j Let α be the comprehensive weight of the j-th dimension feature, α be a preset balancing factor used to adjust the proportion of objective weights, and β be the weight of the j-th dimension feature. jThe subjective weight parameter for the j-th dimension is preset for industry experts. The final comprehensive credit assessment factor is calculated as follows:

[0046]

[0047] Where C is the output comprehensive credit assessment factor, x j This is the feature value of the j-th dimension in the standardized multidimensional original feature vector set. Through the above-mentioned entropy weighting method for objective weighting, expert experience for subjective correction, and weighted fusion processing, the standardized multidimensional original feature vector set generated in the previous step is transformed into a single-dimensional comprehensive credit assessment factor that includes both static credit indicators and dynamic behavioral indicators of enterprises. This achieves a quantitative assessment of the credit status of all parties in cross-enterprise collaborative scenarios, eliminates the bias caused by a single data source, and provides high-confidence input data for the subsequent normalization mapping of trust metrics.

[0048] S1.3: The comprehensive credit assessment factor is processed by numerical interval mapping using a linear normalization algorithm to eliminate the distribution differences between data of different dimensions, so as to generate a standardized initial trust metric value with a value range between zero and one.

[0049] Receive the scalar set of comprehensive credit assessment factors, which includes static credit indicators and dynamic behavioral indicators of enterprises, output from step S1.2. This set serves as the input for linear normalization processing. The assessment factor values ​​of each enterprise are in an unordered range due to differences in the units and the distribution of data sources.

[0050] The algorithm iterates through all sample data in the comprehensive credit assessment factor set. It uses a maximum value search to determine the highest credit assessment score among all participants in the current collaborative scenario, marking it as the global maximum parameter. Simultaneously, it uses a minimum value search to locate the lowest credit assessment score, marking it as the global minimum parameter, thus constructing an extreme value interval reflecting the current data distribution boundary. The difference between the global maximum and global minimum parameters is calculated to generate a denominator benchmark value representing the data's discrete span. If this denominator benchmark value approaches zero, an anomaly handling mechanism is triggered, and a default uniform distribution value is assigned; otherwise, the difference is used as the proportional divisor for the linear mapping. For each enterprise's comprehensive credit assessment factor in the set, a subtraction operation is performed to remove the global minimum parameter, obtaining the enterprise's relative offset from the lowest credit level. The calculated relative offset is divided by the previously generated denominator benchmark value, performing a division operation to eliminate the influence of the original data's dimensions and compressing the value to a standard unit interval. The division results are truncated at the boundaries, with a lower threshold of zero and an upper threshold of one, forcing all calculation results to be constrained within the closed interval [0,1] to prevent spillover effects caused by extreme outliers, thus generating a standardized initial trust metric that strictly conforms to the probability distribution characteristics. The standardized initial trust metric for the i-th participant is calculated using the following linear normalization formula:

[0051] T i =(C i -C min ) / (C max -C min )

[0052] Among them, T i Let C be the standardized initial trust metric for the i-th participant. i C is the comprehensive credit assessment factor for the i-th participant. max C represents the maximum value of the comprehensive credit assessment factor for all participants in the current collaborative scenario. min This represents the minimum value of the comprehensive credit assessment factor for all participants in the current collaborative scenario. Through the aforementioned extreme value extraction, difference calculation, relative offset quantification, and proportional mapping processing methods, the comprehensive credit assessment factors with different dimensions and distribution ranges generated in the previous step are transformed into standardized initial trust metric values ​​with a strictly limited value range between zero and one. This eliminates the numerical scale differences caused by multi-source heterogeneous data, achieves the comparability of cross-enterprise credit data in a unified metric space, and provides standardized numerical input for the attribute initialization of subsequent dynamic trust graph nodes.

[0053] For example, for four participating companies that have joined the supply chain collaboration network in real time (Company D is a multinational trader, Company E is a regional distributor, Company F is a local warehouser, and Company G is a newly joined service provider), the comprehensive credit assessment factors output by the system in S1.2 are 88.5, 62.3, 45.0, and 91.2, respectively. The system first traverses the dataset and identifies the maximum value C. max The value is 91.2 (corresponding to company G), and the minimum value is C. min The value is 45.0 (corresponding to company F). Calculate the denominator baseline value, i.e., 91.2 minus 45.0 equals 46.2. For company D, calculate its relative offset as 88. Subtracting 45.0 equals 43.5, then dividing by 46.2, yields a standardized initial trust metric of approximately 0.9415.

[0054] S1.4: Based on the standardized initial trust metric value, construct a node initialization model for the dynamic trust graph, map each participant to an independent node entity in the graph, and write the standardized initial trust metric value as a core attribute into the node state space to generate a dynamic trust graph node object with an initial trust metric.

[0055] The system receives the standardized initial trust metric value set (ranging from zero to one) and the corresponding participant identity list from step S1.3 as the core input data for initializing the dynamic trust graph nodes. It defines a node data structure model for the dynamic trust graph, including a unique node identifier field, a node type label field, a static attribute set field, and a dynamic state space field. The dynamic state space field is specifically used to store the real-time changing trust metric values. It iterates through each data item in the standardized initial trust metric value set, extracts the identity identifier of the currently processing enterprise, and maps it to the unique identifier field of an independent node in the graph, ensuring that each participant in the cross-enterprise collaborative network has a unique topological position in the graph space. It calls the node instantiation constructor, creating an independent memory object instance for each participant based on the predefined node data structure model. Static features such as the business scope and registered capital obtained from parsing the enterprise qualification certificate are written into the node's static attribute set field, forming the node's inherent feature description.

[0056] The standardized initial trust metric generated in S1.3 is assigned to the initial trust weight attribute in the node's dynamic state space, establishing a strong binding relationship between the node entity and its trust value. This makes the trust metric a core state parameter that can be directly accessed by the graph algorithm. A timestamp attribute is added to each initialized node object to record the generation time of the trust metric value, so as to determine the freshness and validity of the data during subsequent dynamic updates. An adjacency matrix or adjacency list initialization strategy is adopted, and an empty edge connection container is pre-allocated to each node to store the directed edge data generated based on business interaction relationships in subsequent steps. At this time, the edge container is in an unfilled state, retaining only the complete attribute information of the node itself.

[0057] After node object initialization, the system loads historical transaction records, contract signing records, and logistics route data from the supply chain business relationship database to extract direct business interaction relationships between participants. For each pair of participants with business interaction, a directed edge is created, with the direction from the business initiator to the business receiver. The edge data structure includes a unique edge identifier, source node identifier, target node identifier, edge type label (such as "supply relationship", "logistics relationship", "contract relationship"), interaction frequency statistics, recent interaction timestamp, and edge weight attribute. The initial value of the edge weight is determined by a weighted fusion calculation based on the number of historical interactions between the two parties and the average response time; the more interactions and the lower the response latency, the higher the edge weight. All directed edges are verified and written to the corresponding row and column positions in the adjacency matrix, and simultaneously stored in the edge storage area of ​​the graph database, transforming the dynamic trust graph from an isolated set of nodes into a directed weighted graph with a complete topological structure.

[0058] Through the above node model definition, identity mapping, instantiation construction, attribute injection, and state space initialization processing, the discrete standardized initial trust metric value generated in the previous step is transformed into a dynamic trust graph node object with complete data structure and state attributes. This achieves a dimensional improvement from a single numerical indicator to a complex graph entity, laying a solid entity foundation for the subsequent construction of a dynamic trust topology network that reflects cross-enterprise collaborative relationships.

[0059] For example, in a supply chain collaboration scenario, for four participating companies whose standardized initial trust metrics have been calculated (Company A's trust value is 0.943, Company B's is 0.729, Company C's is 0.487, and Company D's is 0.375), the system executes a node initialization process. First, the node data structure is defined, including the following fields: NodeID (string type), Type (enumeration type, such as manufacturer, logistics provider), StaticAttrs (JSON object), TrustScore (floating-point type), UpdateTime (long integer timestamp), and EdgeList (list type). For Company A, the system extracts its identity identifier.

[0060] S1.5: Perform integrity verification and index registration operations on the dynamic trust graph node objects with initial trust metrics, establish a bidirectional association mapping table between node identity and initial trust attributes, and output the initialized dynamic trust graph data structure for subsequent access control processes to call.

[0061] The system receives a set of dynamic trust graph node objects with initial trust metrics generated in step S1.4. This set serves as the input data source for integrity verification and index registration. Each node object contains a unique identifier, a set of static attributes, a dynamic state space, and a timestamp field. It iterates through each instance in the node object set, extracting the node's unique identifier field and the initial trust weight attribute value from the dynamic state space. It performs a non-null value detection logic to verify the completeness of the core data. If a missing identifier or a null trust value is found, an anomaly flag is triggered, and the registration process for that node is paused. For node objects that pass the non-null detection, a data type consistency check is performed. A type reflection mechanism is used to check whether the unique identifier is a string type and whether the initial trust weight is a floating-point type with a value strictly within a closed interval between zero and one. For data items with mismatched types, an automatic conversion operation is performed or a format error log is recorded.

[0062] A node identity index structure based on a hash algorithm is constructed. The unique identifier of a verified node is input into the SHA-256 hash function as the key, generating a fixed-length hash digest which serves as the primary key of the index table. This ensures that the time complexity of node retrieval in cross-enterprise collaborative networks reaches constant level. A bidirectional association mapping table between node identity identifiers and initial trust attributes is established. The first record in the in-memory database maps the hash primary key to the memory address pointer of the node object, enabling rapid location from identity to entity. Simultaneously, a second reverse record maps the memory address pointer of the node object back to the original identity identifier, forming a closed-loop reference relationship to support reverse tracing queries. Node objects that have completed verification and index registration are written to persistent storage. The node data structure is serialized using an append-only write mode and stored in the node storage area of ​​the graph database. The latest mapping snapshot is updated in the index cache layer. Through the above-mentioned non-empty detection, type verification, hash index construction, and bidirectional mapping establishment processes, the discrete dynamic trust graph node objects generated in the previous step are transformed into an initial graph data structure with complete data constraints and efficient retrieval capabilities. This enables standardized access and traceable management for cross-enterprise participants in the trust topology network, providing a highly reliable data foundation for the dynamic matching of subsequent access control policies.

[0063] For example, in a supply chain collaboration scenario, for the four participating companies that have completed node initialization (Company A node ID is "ENT-A-001", trust value 0.943; Company B node ID is "ENT-B-002", trust value 0.729; Company C node ID is "ENT-C-003", trust value 0.487; Company D node ID is "ENT-D-004", trust value 0.375), the system performs an integrity verification and index registration process. First, the four node objects are traversed. The "ENT-A-001" node is checked, confirming its ID field is not empty and is of string type. The trust value 0.943 is a floating-point number within the range [0,1], and the verification passes. The "ENT-B-002" node is checked, and its trust value field is found to be incorrectly stored as the string type "0.729". The system automatically calls the type conversion interface to convert it to the floating-point number 0.729, and the verification passes. The "ENT-C-003" node is checked, and its ID field is found to be empty. The system marks this node as an abnormal state and skips subsequent registration steps, generating an error log "Node ID Missing". The "ENT-D-004" node is checked, and all data conforms to the specifications, and the verification passes. For the three nodes that passed the verification, the system calculates their SHA-256 hash values: the hash key for enterprise A is "a1b2...", for enterprise B it is "c3d4...", and for enterprise D it is "e5f6...". The system constructs a bidirectional mapping table in memory. The first record establishes a mapping relationship: Key="a1b2..." points to memory address 0x7FFA of enterprise A node, and Value=0x7FFA points to "ENT-A-001"; the second record establishes a mapping relationship: Key="c3d4..." points to memory address 0x7FFB of enterprise B node, and Value=0x7FFB points to "ENT-B-002"; the third record establishes a mapping relationship: Key="e5f6..." points to memory address 0x7FFD of enterprise D node, and Value=0x7FFD points to "ENT-D-004". The system then writes the serialized data from these three nodes to the Node storage area of ​​the Neo4j graph database and updates the index snapshot in the Redis cache layer. The process successfully intercepted one abnormal node, automatically repaired one node with a type error, and built an index structure containing three valid nodes. This reduced the response time of the access control module from milliseconds to microseconds when querying the trust status of any enterprise, significantly improving the data consistency and query efficiency of the dynamic trust graph.

[0064] Step S2: Receive the access request initiated by the data requester, generate a zero-knowledge proof credential using local sensitive context information based on the zero-knowledge proof protocol, and verify the compliance of the operating environment without disclosing the specific geographical location and device fingerprint. Specifically, this includes:

[0065] S2.1: Obtain the original geographic location coordinate sequence and the original device fingerprint feature vector collected by the data requesting terminal, perform noise removal processing on the original geographic location coordinate sequence based on the sliding window filtering algorithm, and perform standardization and normalization processing on the original device fingerprint feature vector to output standardized geographic location time series data and standardized device fingerprint feature code.

[0066] S2.2: Receive the standardized geographic location time series data and the standardized device fingerprint feature code, construct a zero-knowledge proof circuit model based on the elliptic curve discrete logarithm problem, map the standardized geographic location time series data as location assertion input variables and map the standardized device fingerprint feature code as device assertion input variables, so as to generate a zero-knowledge proof logic circuit structure containing location compliance constraints and device integrity constraints.

[0067] The system receives geolocation time-series data and device fingerprint features, processed by sliding window filtering and standardization, as the basic input variables for constructing a zero-knowledge proof logic circuit. Based on the mathematical properties of the elliptic curve discrete logarithm problem, a prime-order bilinear pairing group satisfying security parameter requirements is selected as the underlying algebraic structure. Generators and group operation rules are defined to establish the foundation of cryptographic primitives. The latitude and longitude coordinate components in the standardized geolocation time-series data are mapped to location assertion input variables within the circuit, establishing a one-to-one correspondence between coordinate values ​​and elements of a finite field.

[0068] The hardware identifier hash value in the standardized device fingerprint feature code is mapped to the device assertion input variable in the circuit, forming a digital constraint basis for device authentication. A logic gate circuit submodule containing location compliance constraints is constructed, setting geofence boundary threshold parameters, and using comparator logic to determine whether the input coordinates fall within a preset polygonal permission area. A logic gate circuit submodule containing device integrity constraints is constructed, setting a trusted device fingerprint whitelist hash set, and using matching logic to verify whether the input device feature code exists in the authorization set. Arithmetic transformation techniques are used to convert the above location compliance constraints and device integrity constraints into a polynomial equation system, defining constraint polynomials such that the equation system has a solution only if all constraints are simultaneously satisfied. A rank-1 constraint system model is used to structurally encapsulate the transformed polynomial equation system, generating a zero-knowledge proof logic circuit structure consisting of input variables, auxiliary variables, and constraint matrices. By using the circuit modeling and constraint mapping processing method based on the elliptic curve discrete logarithm problem, the standardized environmental perception data output in the previous step is transformed into a zero-knowledge proof logic circuit structure with cryptographic verifiability. This enables the logical pre-definition of operational environment compliance without revealing the original geographical coordinates and device fingerprint plaintext, providing a rigorous mathematical constraint framework for subsequent witness generation and proof ciphertext output.

[0069] S2.3: Based on the zero-knowledge proof logic circuit structure, perform a witness generation operation, and use the Groth16zk-SNARKs proof generation algorithm (using the KZG polynomial commitment scheme and BN254 elliptic curve) to perform a nonlinear arithmetic transformation on the location compliance constraint and the device integrity constraint, so as to output an initial zero-knowledge proof credential data packet containing the ciphertext of the location range validity proof and the ciphertext of the device identity authenticity proof.

[0070] The system receives location compliance constraints and device integrity constraints defined by the zero-knowledge proof logic circuit structure as the logical constraint basis for the witness generation process. It invokes the KZG polynomial commitment scheme in the zk-SNARKs proof generation algorithm to map the linear and multiplicative constraints in the rank-1 constraint system to a system of polynomial equations over a finite field. A private witness vector containing location assertion input variables and device assertion input variables is constructed; this vector is instantiated in memory but its plaintext values ​​are not exposed. An arithmetic transformation operation is performed, using a quadratic arithmetic procedure model to transform each constraint gate in the logic circuit into a corresponding set of polynomial coefficients. For location range validity verification, BN254 elliptic curve scalar multiplication is used to encrypt the location coordinate constraint polynomial, generating location-related intermediate commitment values. For device identity authenticity verification, a homomorphic hiding operation is performed on the hash constraint polynomial corresponding to the SHA-256-based Merkle path verification logic to generate device-related intermediate commitment values. The initial proof data structure is constructed by aggregating location-related and device-related intermediate commitment values ​​using the group element encoding rules required for bilinear pairing on the BN254 curve. The initial proof data structure is blinded by applying a poison parameter from a fixed authoritative trust setting parameter (e.g., the Perpetual Powers of Tau Phase 2 output), eliminating the direct linear correlation between the witness vector and the final proof ciphertext. The quotient polynomial part of the proof polynomial is calculated, and the solution space trajectory satisfying all constraints is determined using Lagrange interpolation. The values ​​of the quotient polynomial at specific evaluation points are extracted and encoded as elliptic curve point coordinates on groups G1 and G2 (G1 and G2 are subgroups on the BN254 curve). The combined encoded elliptic curve point coordinates form an initial zero-knowledge proof credential data packet containing the ciphertext of location range validity proof and the ciphertext of device identity authenticity proof.

[0071] By employing the nonlinear arithmetic transformation based on the Groth16 zk-SNARKs proof generation algorithm and the KZG polynomial commitment processing method, the zero-knowledge proof logic circuit structure and private witness data constructed in the previous step are transformed into an initial zero-knowledge proof credential data packet with cryptographic unforgeability. This enables the mathematical proof of operational environment compliance without disclosing the plaintext geographical coordinates and device fingerprint features, providing a standardized ciphertext input object for subsequent bilinear pairing pre-verification.

[0072] S2.4: Obtain the initial zero-knowledge proof credential data packet, and perform local pre-verification processing on the location range validity proof ciphertext and the device identity authenticity proof ciphertext based on the bilinear pairing verification mechanism to remove invalid proof fragments and output a compliant zero-knowledge proof credential that has passed integrity verification.

[0073] S2.5: Receive the compliance zero-knowledge proof credential that has passed integrity verification, and perform fingerprint extraction processing on the compliance zero-knowledge proof credential that has passed integrity verification based on the hash digest algorithm to generate an immutable environment compliance verification token for subsequent blockchain audit and evidence storage.

[0074] The system receives a compliance zero-knowledge proof credential that has undergone integrity verification as the raw data input for generating an immutable environment compliance verification token. It then extracts the ciphertext of the location range validity proof, the ciphertext of the device identity authenticity proof, and the ciphertext of the overall circuit satisfaction proof from the compliance zero-knowledge proof credential, and constructs a composite data structure to be hashed.

[0075] The SHA3-256 hash digest algorithm is used to initialize the hash state register and set the data block filling rules for the absorption phase. The composite data structure is then divided into blocks of fixed byte length. A permutation function is performed on each data block, and the data block content is mixed into the hash state through an XOR operation, updating the state matrix to reflect the fingerprint characteristics of the current input data. After all data blocks have been absorbed, the squeezing phase begins. A fixed-length digest output value is extracted from the final state matrix to generate a unique digital fingerprint representing the environmental compliance of this access request. The generated digital fingerprint is serialized and concatenated with timestamp information, requester identifier, and blockchain smart contract address to form standardized token payload data. A hash operation is then performed on the token payload data again to generate an environmental compliance verification token with collision resistance. This token serves as the unique index key for subsequent on-chain notarization.

[0076] By using the fingerprint extraction and token construction processing method based on the hash digest algorithm, the compliance zero-knowledge proof certificate output by the previous step, which has undergone integrity verification, is transformed into an environment compliance verification token with cryptographic uniqueness and immutability. This enables efficient indexing and secure anchoring of cross-enterprise data access records in blockchain audit scenarios, significantly improving the accuracy and audit efficiency of permission change tracing.

[0077] Step S3: Real-time monitoring of order status changes, logistics node updates, and anomaly alarm event signals in the supply chain business flow; inputting these event signals as weighting factors into the dynamic trust graph; calculating and outputting the real-time updated dynamic trust score. Specifically, this includes:

[0078] S3.1: Perform protocol parsing and standardization cleaning on order status change logs, logistics node update messages and abnormal alarm signals in the supply chain business flow to eliminate data format heterogeneity and extract standard business event sequences containing timestamps, event types and associated entity identifiers.

[0079] S3.2: Based on the standard business event sequence, semantic matching and risk level labeling are performed using a predefined risk assessment mapping rule base to transform discrete event types into initial event weight vectors with positive and negative directionality and intensity quantification.

[0080] The construction method for the risk assessment mapping rule base is as follows:

[0081] The risk assessment mapping rule base is an offline pre-built knowledge base. Its construction process includes three main stages: risk pattern extraction, mapping relationship labeling, and rule base encapsulation.

[0082] Risk pattern extraction phase

[0083] Using historical audit reports, industry security standards (such as ISO 28000), and typical risk event case libraries as raw corpora, a text mining method based on dependency parsing is employed to extract frequently occurring event description patterns and their corresponding risk outcome descriptions from unstructured text. Specific operations include: segmenting and tagging the text corpus with words, identifying trigger words (such as action verbs like "delay," "overdue," "non-conforming," "early," and "timely") and their associated objects (such as business entities like "orders," "logistics," "quality inspection," and "payment"), and grouping expressions with the same semantic core into the same event type. Through this process, a limited set of event types covering core supply chain risk areas such as delivery timeliness, quality compliance, information security, contract performance, and payment settlement is formed, with each event type assigned a unique code identifier.

[0084] Mapping relationship calibration stage

[0085] This is the core step in building the rule base, responsible for establishing the deterministic correspondence between event types and weight vector components. The specific method is as follows:

[0086] (1) Directional Judgment Rules: For each event type, it is classified into two categories based on its actual impact on the trust relationship between supply chain entities. The specific judgment principle is as follows: event types that reflect improved performance capabilities, enhanced collaboration efficiency, or proactive assumption of additional responsibilities by entities are classified into the positive enhancement category, and the directional mark is recorded as positive; event types that reflect default, dereliction of duty, information concealment, or losses to upstream and downstream entities are classified into the negative weakening category, and the directional mark is recorded as negative. This judgment is made by domain experts based on general business common sense in supply chain management to ensure that the marking results have consensus within the industry.

[0087] (2) Benchmark Risk Score Determination Method: The Delphi method was used to organize multiple supply chain risk management experts to independently score the potential damage or gain to business continuity for each event type. A five-point scale was used for scoring, with higher scores indicating a stronger impact of the event on trust assessment. After summarizing the scores from all experts, the median and interquartile range were calculated. Outliers were removed, and the arithmetic mean was taken as the benchmark risk score for that event type to ensure the stability of the score and consistency among experts.

[0088] (3) Generation of mapping relationship table: Each event type code, directional identifier, and baseline risk score obtained in the above steps is combined into a mapping record, and all records are summarized to form the core mapping relationship table of the rule base. This table uses the event type code as the unique primary key and supports O(1) complexity hash lookup. The table structure contains five fields: event type code, event type name, directional identifier, baseline risk score, and risk domain label.

[0089] (4) Preset dynamic correction parameters: To further improve the context adaptability of the mapping, a set of optional correction factors is pre-associated for each mapping record, including enterprise historical compliance correction factors and business urgency correction factors. The values ​​of the correction factors are pre-calibrated in the range of 0.5 to 1.5. During actual operation, the correction factors are dynamically selected and applied to the benchmark risk score based on the context information accompanying the event.

[0090] Rule base encapsulation and version management stage

[0091] The mapping relationship table is serialized and stored according to a preset JSON data structure to form a machine-readable rule base file. The rule base supports version number management and effective timestamp recording, and each revision retains historical versions for rollback. The admission of new event type mapping records requires historical backtracking verification—that is, the new mapping rule is applied to a historical event dataset with known results, its risk assessment accuracy is calculated, and it can only be merged into the main version if the accuracy reaches a preset threshold.

[0092] For the standard business event sequence output from step S3.1, which includes timestamps, event types, and associated entity identifiers, a deep semantic parsing operation is performed by calling a pre-built risk assessment mapping rule base. Entity recognition algorithms from natural language processing are used to extract key action verbs and object objects from the event description text, transforming the unstructured event description into a standardized semantic triple structure. The extracted semantic triples are then fuzzily matched with risk pattern templates in the rule base to determine the risk category label for each standard business event. For successfully matched risk categories, the baseline risk score and directional coefficient defined in the rule base are retrieved, where the directional coefficient indicates whether the event positively strengthens or negatively weakens the trust relationship. Based on the contextual parameters of the event, including the historical compliance records of the involved enterprises and the current business urgency, the baseline risk score is dynamically adjusted. A weighted summation method is used to multiply the adjusted baseline score by the directional coefficient to generate an initial event weight vector representing the magnitude and nature of the impact of a single event.

[0093]

[0094] Where W is the calculated initial event weight vector, S base The baseline risk score is retrieved from the rule base, D is the directional coefficient with a value range of [-1,1], α is the environmental sensitivity adjustment factor, and C is the comprehensive context environmental parameter value.

[0095] S3.3: The initial event weight vector is weighted and fused by element-wise multiplication of time decay coefficient and behavior frequency statistical factor to generate dynamic event influence coefficient that can reflect recent behavior trends and suppress historical noise interference.

[0096] For the initial event weight vector output from step S3.2, which has positive and negative directionality and intensity quantification, a multi-dimensional weighted fusion calculation is performed using a time decay function based on timestamps and a frequency factor based on historical behavior statistics. The timestamp parameter bound to the initial event weight vector is extracted, and the time difference between the current system time and the event occurrence time is calculated. This time difference is substituted into a preset exponential decay model to generate a time sensitivity coefficient. An exponential function with the natural constant e as the base is used to construct the time decay kernel, and a half-life parameter is set to control the rate of decline of the influence of historical events over time, ensuring that recent supply chain anomalies or compliance behaviors have a dominant weight in trust assessment. Frequency statistics are performed on similar events occurring within a sliding time window for the same entity identifier, and the behavior density per unit time is calculated and input as a behavior frequency statistical factor into a nonlinear saturated growth function. A hyperbolic tangent function is used to map the behavior frequency, suppressing drastic fluctuations in trust scores caused by high-frequency, single-type events in a short period, and preventing trust assessment distortion caused by malicious order manipulation or concentrated error reporting. The generated time sensitivity coefficient and behavior frequency statistical factor are multiplied element-wise with the initial event weight vector obtained in the previous steps to synthesize a dynamic event influence coefficient K that represents recent behavioral trends and filters out historical noise interference.

[0097]

[0098] Where K is the calculated dynamic event influence coefficient, W is the initial event weight vector output from the previous step, λ is the time decay rate constant, Δt is the time difference between the current time and the event occurrence time, β is the frequency sensitivity adjustment coefficient, and F is the behavior frequency statistics within the sliding window.

[0099] The time decay rate constant λ is set to a fixed value of 0.01 (unit: seconds). -1 The corresponding half-life is approximately 69 seconds, meaning the influence of historical events decays at this rate over time. Historical events older than 300 seconds have a decay coefficient below 0.05, considered essentially extinguished. The sliding time window is fixed at 300 seconds (5 minutes), sliding forward in 60-second increments. At each increment, the system counts the number of occurrences F of similar events for the same subject within the past 300 seconds. The frequency sensitivity coefficient β is fixed at 0.5, ensuring that the saturation value approaches 1 when F reaches 8 times, effectively suppressing drastic fluctuations in trust scores caused by high-frequency abnormal behavior. The components of the initial event weight vector W have undergone dimensional normalization before being input into this step, with each component's value falling within the range of [-1, 1].

[0100] By using the weighted fusion processing method of time decay and frequency suppression described above, the static initial event weight vector generated in the previous step is transformed into a dynamic event influence coefficient that can reflect the latest behavioral trend of the subject in real time and effectively suppress historical noise interference. This achieves accurate quantification of the dynamic evolution of trust relationships in the cross-enterprise collaboration process of the supply chain, and provides a time-sensitive and noise-resistant input data foundation for the subsequent iterative update of node attributes under the graph propagation algorithm.

[0101] S3.4: Based on the dynamic event impact coefficient, locate the corresponding enterprise user node and its adjacent edges in the dynamic trust graph, and perform node attribute iterative update processing under the graph propagation algorithm to calculate the real-time dynamic trust score that integrates the global topology and local behavior feedback.

[0102] S3.4: The dynamic event impact coefficients output in step S3.3 are used to perform node location and adjacency retrieval operations within the pre-constructed dynamic trust graph data structure. The associated entity identifiers bound to the dynamic event impact coefficients are extracted and used as primary key indexes for hash lookups in the graph node set to accurately pinpoint the target enterprise user node where the behavioral event occurred. The current attribute vector of the target enterprise user node is read, containing historical accumulated trust values, reputation level tags, and the most recent update timestamp, serving as the baseline state data for this iteration. The edge set of the dynamic trust graph is traversed, retrieving all adjacent nodes directly connected to the target enterprise user node and their connecting edge weights, constructing a local topological subgraph to capture the trust transmission path between upstream and downstream enterprises. The graph propagation algorithm employs a personalized PageRank algorithm, transforming the dynamic event impact coefficients calculated in previous steps into incremental correction terms for node attributes. Convergence thresholds and maximum iteration count parameters are set, the node trust score vector is initialized, and the dynamic event impact coefficients are superimposed on the initial score of the target node, forming an intermediate state vector with immediate behavioral feedback. Matrix multiplication is performed, and the intermediate state vector of the target node is probabilistically diffused to its first-order and multi-order neighboring nodes using the transpose of the adjacency matrix and the normalized edge weight matrix. During the diffusion process, a damping factor is introduced to simulate the natural decay effect of trust relationships in long-chain transmission, preventing trust scores from accumulating infinitely in acyclic paths. For each neighboring node, the propagation contribution values ​​from all in-degree neighboring nodes are aggregated, and the weighted sum is calculated as the new trust score estimate for that node. The newly calculated trust score estimate is compared with the historical score from the previous iteration. If the change in scores for all nodes is less than a preset convergence threshold, a steady state is reached and the iteration terminates; otherwise, the new score is updated to the current state, and the next propagation cycle begins. After iterative convergence, the final attribute value of the target enterprise user node is extracted. This value integrates the impact of local real-time behavioral shocks and the global network topology, generating a real-time dynamic trust score.

[0103]

[0104] Among them, T (k+1) Let M be the node trust score vector after the (K+1)th iteration, d be the damping factor ranging from 0 to 1, and M be the node trust score vector. T T is the transpose of the adjacency matrix after column normalization. (k) Let V be the trust score vector for the k-th iteration, and let V be the personalized reset vector containing the dynamic event impact coefficient, V=[K1,K2,...,K]. n ], where K i The dynamic event impact coefficient is assigned to each node. Through the node attribute iterative update processing method under the graph propagation algorithm described above, the discrete dynamic event impact coefficients generated in the previous step are transformed into real-time dynamic trust scores that integrate global topology and local behavioral feedback. This enables rapid transmission and accurate quantification of trust relationships in cross-enterprise collaboration scenarios, providing a highly reliable decision-making basis for the dynamic adjustment of subsequent access control strategies.

[0105] Step S4: Based on the numerical range of the dynamic trust score, match the corresponding data anonymization algorithm strength level and field visibility range strategy to generate a fine-grained access control parameter group containing masking rules or aggregation granularity. Specifically, this includes:

[0106] S4.1: Obtain the real-time updated dynamic trust score output by the dynamic trust graph, and standardize the dynamic trust score to eliminate the influence of different units and generate a standardized trust metric.

[0107] Receives the raw dynamic trust score output in real time from the dynamic trust graph. This score represents the comprehensive trust measurement result of the requester in the current time window in a cross-enterprise collaboration scenario. It calls the preset score normalization algorithm processing module to read the dimensionless parameter set stored in the configuration register, including the historical statistical mean and standard deviation boundary values ​​of the weights of each trust dimension. It performs a linear transformation operation on the raw dynamic trust score and calculates the standardized intermediate variables using the following formula:

[0108]

[0109] Where X is the original dynamic trust score input, μ is the historical statistical mean of the corresponding trust dimension, σ is the standard deviation of the corresponding trust dimension, and Z is the standardized intermediate variable after eliminating the influence of dimensions.

[0110] The calculated standardized intermediate variables are mapped to a preset confidence interval. A truncation function is used to handle abnormal fluctuations exceeding the threshold, ensuring that the standardized trust metric falls within a closed interval of zero to one. The truncated values ​​are then precision aligned, retaining four decimal places. A metadata structure containing a timestamp, requester identifier, and standardized trust metric is generated, serving as direct input for dynamic trust level determination. Through the above score normalization and anomaly truncation methods, the multi-source heterogeneous dynamic trust scores output in the previous step are transformed into a standardized trust metric with unified dimensions, achieving the expected technical effect of eliminating differences in the dimensions of different evaluation dimensions and improving the accuracy of subsequent permission matching.

[0111] S4.2: Receive the standardized trust metric and map it to a predefined trust level sequence to determine the dynamic trust level label of the current requester. Receive the metadata structure generated in the previous step, containing a timestamp, requester identifier, and standardized trust metric, and extract the standardized trust metric as the core input variable for the interval mapping operation. Call the pre-set multi-dimensional interval partitioning logic module to read the trust level sequence definition table stored in non-volatile memory. This table contains multiple mutually exclusive and continuous numerical sub-intervals and their corresponding discretized level labels. Compare the standardized trust metric with the preset lower and upper threshold values ​​of the interval sequentially. When a standardized trust metric is detected to be greater than or equal to the lower threshold of a certain level and less than the upper threshold of that level, it is determined that the metric falls within the current confidence interval range. If the standardized trust metric is exactly equal to the upper threshold of the highest level, it is classified as the highest trust level through closed interval inclusion logic. For values ​​at the critical points of two adjacent levels, an upward rounding strategy is used to avoid misjudgment of levels due to floating-point precision errors, ensuring that low-score requests are not incorrectly promoted to high levels.

[0112] Based on the successfully matched interval index, a unique dynamic trust level label is retrieved from the trust level sequence definition table. This label uses an enumeration type encoding to represent the access permission benchmarks for different levels. The determined dynamic trust level label is written to a temporary register and associated with the original requester identifier, forming an intermediate state object with a level attribute. Through the above multi-dimensional interval partitioning and boundary matching processing, continuously changing standardized trust metrics are transformed into discrete dynamic trust level labels, realizing a logical mapping from quantitative values ​​to qualitative permission levels, providing a clear decision-making basis for the precise invocation of subsequent differentiated de-identification strategies.

[0113] For example, in an electronic component supply chain collaboration scenario, the system receives an access request from a Tier 3 packaging and testing plant. The standardized trust metric calculated in the previous steps is 0.7245. The system loads a predefined trust level sequence definition table, which divides the closed interval from zero to one into five levels:

[0114] Level L1 corresponds to the interval from 0.0 to 0.2;

[0115] Level L2 corresponds to the interval from 0.2 to 0.4;

[0116] Level L3 corresponds to the interval 0.4 to 0.6;

[0117] Level L4 corresponds to the interval 0.6 to 0.8;

[0118] Level L5 corresponds to the interval from 0.8 to 1.0.

[0119] The system executes boundary value comparison logic. First, it checks if 0.7245 is greater than or equal to 0.0 and less than 0.2, which is negative. Next, it checks if it is greater than or equal to 0.2 and less than 0.4, which is also negative. It then checks if it is greater than or equal to 0.4 and less than 0.6, which is still negative. Finally, it checks if it is greater than or equal to 0.6 and less than 0.8. At this point, 0.7245 satisfies both the condition of being greater than or equal to 0.6 and the condition of being less than 0.8, so the logic is valid. The system locks the value into the L4 level range and retrieves the corresponding dynamic trust level label from the definition table.

[0120] The construction and retrieval process of the dynamic trust level label table is as follows:

[0121] Standardized trust metric acquisition: Extract the standardized trust metric from the metadata structure generated in the previous steps. This metric is a continuous floating-point number with a value range of [0,1], and serves as the sole input variable for the range mapping operation.

[0122] Trust level sequence definition table loading: The system loads a pre-built trust level sequence definition table from non-volatile memory. This table divides the closed interval [0,1] into five mutually exclusive and consecutive numerical subintervals. Each subinterval is associated with a unique discretized level label. The level labels are encoded using an enumeration type L1 to L5, representing the increasing trust level from low to high.

[0123] Interval Boundary Comparison and Classification Determination: The standardized trust metric is sequentially compared with the lower and upper thresholds of each interval in the definition table. The comparison uses a closed-open interval mathematical convention, determining whether the metric value is greater than or equal to a certain lower threshold and less than the upper threshold of that interval. When this condition is met, the metric is determined to fall into the current interval, and a corresponding dynamic trust level label is generated. If the metric value equals the upper threshold of the highest level (i.e., 1.0), it is directly classified into L5 level using closed interval inclusion logic, and a corresponding dynamic trust level label is generated. For any input value, the above comparison logic ensures that only one interval determination is valid across the five levels, guaranteeing the uniqueness of the mapping results; that is, each standardized trust metric uniquely corresponds to one dynamic trust level label.

[0124] Retrieval and binding of dynamic trust level tags: Based on the successfully matched interval index, retrieve the corresponding enumerated type dynamic trust level tag from the definition table. Associate and bind this dynamic trust level tag with the requester identifier to form an intermediate state object with a level attribute.

[0125] S4.3: Read the dynamic trust level label, call the pre-built de-identification algorithm strength level configuration table, and filter out the target de-identification algorithm strength level corresponding to the dynamic trust level label.

[0126] The de-identification algorithm strength level configuration table uses an integer value between 1 and 10. A smaller value indicates weaker de-identification (retaining more original information), while a larger value indicates stronger de-identification (hiding more original information). Specific mapping examples: Trust level L5 (highest trust level) corresponds to de-identification strength level 1 or 2, indicating almost no de-identification; Trust level L4 corresponds to de-identification strength level 3 or 4; Trust level L3 corresponds to de-identification strength level 5 or 6; Trust level L2 corresponds to de-identification strength level 7 or 8; Trust level L1 (lowest trust level) corresponds to de-identification strength level 9 or 10, indicating the highest level of de-identification. This mapping relationship is pre-stored in the configuration table and can be dynamically adjusted by operations and maintenance personnel according to security policies.

[0127] The system receives an intermediate state object generated in the preceding steps, containing the requester identifier and a dynamic trust level label, and extracts the dynamic trust level label. A fast retrieval operation is performed on the extracted dynamic trust level label to locate the specific row address of the de-identification algorithm strength level configuration table stored in non-volatile memory. A predefined mapping relationship record is read from this configuration table, which establishes a one-to-one correspondence between discrete trust level labels and continuous de-identification strength coefficients. The read configuration data is parsed to separate the target de-identification algorithm type identifier and its corresponding strength adjustment parameter value specified for the current trust level. If a multi-level cascading strategy exists in the configuration table, multiple candidate de-identification algorithms are weighted and fused according to preset priority weights to generate a comprehensive target de-identification algorithm strength level. The selected target de-identification algorithm strength level is encapsulated into a standardized data structure, with an effective timestamp and strategy version number, forming a strategy instruction set that can be directly called by the subsequent field visibility range calculation module.

[0128] By using the above-mentioned hash index-based strategy retrieval and parameter parsing processing method, the qualitative dynamic trust level label determined in the previous step is transformed into a quantitative target de-identification algorithm strength level, realizing a precise mapping from abstract security level to specific execution parameters. This ensures that requesters with different trust levels can obtain data protection strength that matches their risk status in real time, significantly improving the dynamic adaptability and execution efficiency of access control policies in cross-enterprise collaboration scenarios.

[0129] S4.4: Load the target desensitization algorithm strength level, combine it with the sensitivity metadata of the supply chain data fields, and generate a set of masking rules and aggregation granularity for specific data fields.

[0130] The system receives the target de-identification algorithm strength level identifier and supply chain data field sensitivity metadata set output from the previous steps, and extracts the strength coefficient values ​​as the basic weight parameters for mask calculation. It then parses the data field sensitivity metadata to identify the sensitivity level label, data type attribute, and business context relevance of each field to be accessed, constructing a structured input vector containing field identifiers, original data lengths, and sensitivity factors. Finally, it logically assembles the generated bitmasking rules, substring replacement patterns, and aggregation dimension parameters for each field to form a masking rule set and aggregation granularity definition list covering all requested fields.

[0131] S4.5: Integrate the mask rule set with the aggregation granularity, and construct a structured fine-grained access control parameter group based on the parameter encapsulation protocol, which serves as the direct execution basis for subsequent data differentiation processing.

[0132] Step S5: Based on the fine-grained access control parameter group, differentiate the target supply chain data. If the dynamic trust score is higher than the preset high trust threshold, output the complete plaintext data; if it is lower than the preset low trust threshold, output the trend data after key value masking or aggregation. Specifically, this includes:

[0133] S5.1: Obtain the dynamic trust score and preset high trust threshold and preset low trust threshold. Use an interval mapping method to perform logical judgment processing on the dynamic trust score to generate an access permission level label containing a complete access identifier, a partial access identifier, or an access denied identifier. Obtain the dynamic trust score and preset high trust threshold and preset low trust threshold as the basic input data for access permission level determination.

[0134] The interval mapping module is invoked, and the dynamic trust score is input as an independent variable into the segmented logic decision function. A three-dimensional state space is constructed, comprising high-trust, medium-trust, and low-trust intervals. The high-trust interval is defined as the range where the dynamic trust score is greater than a preset high-trust threshold; the medium-trust interval is defined as the range where the dynamic trust score is between the preset high-trust threshold and the preset low-trust threshold; and the low-trust interval is defined as the range where the dynamic trust score is less than the preset low-trust threshold. A comparison operator is used to perform an initial comparison between the dynamic trust score and the preset high-trust threshold. If the dynamic trust score is strictly greater than the preset high-trust threshold, a complete access identifier generation instruction is triggered. A second comparison operator is used to perform a second comparison between the dynamic trust score and the preset low-trust threshold. If the dynamic trust score is strictly less than the preset low-trust threshold, an access denial identifier generation instruction is triggered. If the dynamic trust score does not trigger either of the above two extreme cases (i.e., it falls within the intermediate value range), a partial access identifier generation instruction is triggered. Using an enumeration type definition method, the complete access identifier is encoded as the highest priority permission code, the partial access identifier is encoded as the medium priority permission code, and the access denial identifier is encoded as the lowest priority permission code. The generated permission code is written into the header field of the access permission level label data structure through bitwise operations. A hash verification algorithm is used to verify the integrity of the generated access permission level label, ensuring that the label content has not been tampered with and that the format conforms to the protocol specifications. Through interval mapping, the dynamic trust score obtained in the previous step is converted into an access permission level label containing a complete access identifier, a partial access identifier, or an access denied identifier, achieving the desired technical effect of dynamically adapting fine-grained access control policies based on real-time trust status.

[0135] For example, in a cross-enterprise supply chain collaboration scenario, the system receives a dynamic trust score of 0.85 from a logistics service provider node, with a preset high trust threshold of 0.80 and a preset low trust threshold of 0.40. The system calls the interval mapping module, first performing an initial size comparison to determine if 0.85 is greater than 0.80. If the result is true, a complete access identifier generation instruction is triggered. The system encodes the complete access identifier as the hexadecimal value 0x01, representing the highest priority permission code. If another participating node has a dynamic trust score of 0.65, the system performs an initial size comparison to determine if 0.65 is greater than 0.80 (false), then performs a second size comparison to determine if 0.65 is less than 0.40 (false), determining that the score falls into the medium trust interval. A partial access identifier generation instruction is then triggered, encoding it as the hexadecimal value 0x02, representing the medium priority permission code. If the dynamic trust score of the third-party node is 0.25, the system performs a secondary size comparison to determine that 0.25 is less than 0.40, which is true. This triggers an access denial flag generation instruction, encoding the value as hexadecimal 0x00, representing the lowest priority permission code. The system writes the generated permission code 0x01, 0x02, or 0x00 into the header field of the corresponding requester's access permission level label, and calculates the SHA-256 hash value, appending it to the end of the label. For a dynamic trust score of 0.65, the generated access permission level label explicitly indicates that subsequent steps can only extract key numerical masking rules and time window aggregation granularity parameters from the cost details field, and cannot obtain the original plaintext data. This process significantly improves the efficiency of permission determination in cross-organizational data interaction, ensuring that subjects with different trust levels can only access data granularity matching their trust status. This effectively avoids the risk of highly sensitive data being leaked to low-trust subjects, while also eliminating the business collaboration rigidity problem caused by static permission configuration.

[0136] S5.2: Receive access permission level labels and the original field set of target supply chain data, and extract key numerical masking rules and time window aggregation granularity parameters for the cost details field and logistics trajectory field based on some access identifiers in the access permission level labels.

[0137] Receive the access permission level label and the target supply chain data raw field set generated in step S5.1 as the data processing input source for this step. Parse the permission code bit field in the header field of the access permission level label to identify the medium priority permission code status that is identified as a partial access identifier.

[0138] Based on the identified partial access identifier status, an index query key is constructed for the dynamic masking rule base. This key contains the requester's subject ID, data type identifier, and current timestamp information. The dynamic masking rule base interface is called, and the constructed index query key is used to retrieve the pre-stored rule mapping table to locate the masking strategy entry for the cost detail field that matches the current business scenario. Key numerical masking rule parameters are extracted from the matched strategy entry, including the mask start offset, mask length in bits, and mask padding character encoding value. The time window aggregation granularity parameter for the logistics trajectory field is then parsed from the same strategy entry to obtain the sliding window size setting and step interval time unit. For the cost detail field, a bit replacement processing logic object is instantiated, and the extracted key numerical masking rule parameters are loaded. For the logistics trajectory field, a spatiotemporal sequence aggregation calculation object is instantiated, and the extracted time window aggregation granularity parameters are loaded to initialize the window boundary of the moving average algorithm. By using dynamic desensitization rule base invocation and parameter extraction processing, the access permission level labels generated in the previous step are transformed into a specific set of execution instructions containing key numerical mask rules and time window aggregation granularity parameters, thereby achieving dynamic adaptation of fine-grained data desensitization strategies to the expected technical effect for subjects with different trust levels.

[0139] S5.3: Use key numerical masking rules to perform bit substitution processing on sensitive numerical data in the original field set of the target supply chain data, and use time window aggregation granularity parameters to perform moving average calculation processing on spatiotemporal series data to generate an intermediate desensitized dataset containing masked numerical fields and aggregated trend fields.

[0140] This step receives key numerical masking rule parameters and time window aggregation granularity parameters extracted from step S5.2, along with sensitive numerical data and spatiotemporal sequence data from the original field set of the target supply chain data, as the core processing objects. For sensitive numerical data fields such as cost details, bit-level data processing is instantiated, loading the start bit offset, mask length bits, and padding character encoding values ​​from the key numerical masking rules to construct the underlying bit operation logic. The original numerical data is converted to a fixed-length binary two's complement representation, and the start position of the bit segment to be processed is located based on the start bit offset. A continuous bit sequence of a specified mask length is extracted, and a bitwise OR operation is used to fuse the extracted bit sequence with the mask template generated by the padding character encoding values, achieving precise replacement of sensitive information bits. The binary data stream after bit replacement is converted back to decimal numerical format, generating a masked numerical field containing asterisks or other masking symbols, ensuring that the original precise numerical value cannot be irreversibly restored. For spatiotemporal sequence data fields such as logistics trajectories, a sliding window calculation module is instantiated, loading the sliding window size setting and step interval time unit from the time window aggregation granularity parameters. A set of historical sampling points that meet the sliding window size setting before the current time point is extracted from the original spatiotemporal sequence, and historical data exceeding the time boundary is removed to maintain the window's real-time performance. A weighted average operation is performed on the discrete latitude and longitude coordinate point sequence within the window, and the aggregated trend coordinate points are calculated using the following formula:

[0141]

[0142] Where P is the aggregated trend coordinate point, and C i Let w be the latitude and longitude coordinate vector of the i-th sampling point within the window. i Here, n is the time decay weighting factor for the corresponding sampling points, and n is the total number of valid sampling points within the window. The initial boundary of the sliding window is moved according to the step interval time unit, and the above weighted average operation is repeated to generate a smoothed spatiotemporal trajectory trend sequence. The generated masked numerical fields and the aggregated trend coordinate sequence are then structurally encapsulated, and field type identifiers and desensitization strategy version number metadata are added to form an intermediate desensitized dataset.

[0143] S5.4: Based on the complete access identifier or intermediate de-identified dataset in the access permission level label, execute the conditional branch routing strategy. If it is a complete access identifier, directly pass through the original field set of the target supply chain data. If it is a partial access identifier, output the intermediate de-identified dataset.

[0144] Receive the intermediate de-identified dataset generated in step S5.3 and the access permission level label output in step S5.1 as the input source for the conditional branch routing strategy in this step. Parse the permission code bit field in the header field of the access permission level label and extract the enumerated values ​​that identify the access level to determine the trust status category of the current requester.

[0145] S5.5: Encapsulate metadata header information for the final delivery data payload and perform integrity check code generation. Push the encapsulated final delivery data payload to the data requester's terminal interface to complete a fine-grained data differentiation delivery process based on dynamic trust scores.

[0146] Step S6: The access request details, zero-knowledge proof verification results, dynamic trust score change trajectory, and final data delivery format are hashed to generate immutable audit evidence data and written into the blockchain ledger. Specifically, this includes:

[0147] S6.1: Perform field-level standardization cleaning and structured encapsulation on access request detail messages, zero-knowledge proof verification credentials, dynamic trust score change sequences, and final data delivery form identifiers to generate original data packets to be stored containing complete operational context, ensuring the format consistency and integrity of multi-source heterogeneous audit data.

[0148] S6.2: Based on the timestamp sequence and participant identity identifiers in the original data packet to be stored, perform lexicographical sorting and deduplication verification to eliminate data redundancy caused by concurrent writing and generate an ordered audit data stream, providing a deterministic input data sequence for subsequent cryptographic digest calculation.

[0149] S6.3: The national cryptographic SM3 hash algorithm is used to perform one-way cryptographic digest operation on the ordered audit data stream to generate a fixed-length unique data fingerprint feature value. This ensures that any minor tampering with the original audit data will result in a significant change in the data fingerprint feature value, thereby achieving anti-counterfeiting protection.

[0150] S6.4: Construct a Merkle tree node structure based on the unique data fingerprint feature value and the hash value of the previous block, and perform secondary hash iteration processing to generate the current block header hash index containing forward and backward linking relationships, forming a logically continuous and physically distributed chain data structure to ensure the irreversibility of historical records.

[0151] The system receives the unique data fingerprint feature value generated in the previous step and the hash value of the previous block header stored in the blockchain ledger as the core input parameters for constructing the current block. The binary string of the unique data fingerprint feature value is extracted as the leaf node data payload of the Merkle tree, ensuring the original integrity of the audit data is mapped to the bottom layer of the tree structure. The system calls the previous block header hash value reading interface to retrieve the 256-bit parent block hash index from the latest state database of the distributed ledger, establishing the logical predecessor position of the current block in the chain topology. The leaf node data payload and the parent block hash index are concatenated according to a preset block header data structure template to form the original block header message to be hashed, containing a timestamp, version number, and difficulty target. The system performs the first SM3 hash operation on the original block header message to be hashed, generating the intermediate-level Merkle tree root node hash value, which represents the aggregate characteristics of all audit transactions in the current batch. A secondary hash iteration input vector is constructed based on the Merkle tree root node hash value and the parent block hash value at the intermediate level. This simulates the double hash verification logic in the proof-of-work mechanism to enhance collision resistance. A second SM3 hash operation is performed, transforming the secondary hash iteration input vector into a fixed-length current block header hash index. This index simultaneously contains the summary information of the current transaction set and the inheritance relationship of the historical chain. The current block header hash index is verified to meet the preset leading zero-bit difficulty constraint. If not, the Nonce field in the block header is adjusted, and the secondary hash iteration process is repeated until it conforms to the consensus rules. The finally generated compliant current block header hash index is written into the header area of ​​the new block, establishing a physical pointer to the parent block, forming a logically continuous and physically distributed chained data structure. Through Merkle tree node construction and secondary hash iteration processing, the single data fingerprint generated in the previous step is transformed into a block header hash index with forward and backward linkage relationships. This achieves structured solidification and irreversible guarantee of audit history in cross-enterprise collaborative scenarios, providing a cryptographic anchor for subsequent smart contract consensus verification.

[0152] S6.5: Based on the current block header hash index, call the smart contract interface to execute consensus verification and distributed broadcast processing, so as to permanently write the new block data containing the complete audit chain into the blockchain ledger storage area, realizing transparent evidence storage and traceable query of the full life cycle permission operation records in cross-enterprise collaborative scenarios.

[0153] Step S7: Monitor the characteristics of unauthorized access patterns in the audited and stored data. If potential risky behavior is detected, trigger the smart contract to execute a trust score freeze instruction. Specifically, this includes:

[0154] S7.1: Perform real-time scanning of the audit evidence data stored in the blockchain ledger, and use sequence pattern mining algorithms to extract unauthorized access pattern feature vectors containing features of high-frequency access during non-working hours and attempts to log in from different locations, in order to construct an initial risk behavior feature set for risk assessment.

[0155] Real-time scanning of audit evidence data stored in the blockchain ledger yields a structured log stream containing user identity identifiers, access timestamps, operation types, and source IP addresses, serving as the foundational data source for unauthorized access pattern mining. A sliding time window mechanism is employed to segment the structured log stream, dividing the continuous time series into fixed-length sub-sequence segments to capture behavioral clustering features within local time domains. A generalized suffix tree construction algorithm is used to index the segmented sub-sequences, and the frequency and distribution density of high-frequency access commands during non-working periods are statistically analyzed by traversing the suffix tree nodes to identify abnormal time-series patterns deviating from normal business rhythms. For the identified abnormal time-series patterns, the ratio of geographical jump distance to time interval for remote login attempts is extracted, and a spatial-temporal violation index is calculated to quantify the physical impossibility of user behavior trajectories. The abnormal time-series pattern feature vector is multi-dimensionally concatenated with the spatial-temporal violation index to construct an initial risk behavior feature set containing temporal violation intensity, spatial jump anomaly degree, and outlier values ​​of operation frequency. By using sequence pattern mining, the raw audit logs from the previous step are transformed into structured initial risk behavior feature set data, enabling automated feature extraction and quantitative characterization of covert unauthorized access behaviors in cross-enterprise collaboration scenarios, providing a high-confidence input basis for subsequent smart contract risk assessment.

[0156] S7.2: Based on the initial risk behavior feature set, a trust score freeze instruction is generated using the smart contract automatic execution logic judgment mechanism to forcibly lock the target user's dynamic trust score to a frozen state value.

[0157] The supply chain data sharing method for cross-enterprise collaboration provided in this application can be applied to, for example... Figure 4 In the application environment shown, terminal 102 communicates with server 104 via a network. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices, and server 104 can be a standalone server or a server cluster consisting of multiple servers.

[0158] In one embodiment, a supply chain data sharing system for cross-enterprise collaboration is provided, comprising:

[0159] Dynamic Trust Graph Module: Acquire enterprise qualification data, historical performance records and third-party credit scores of each participant in cross-enterprise collaboration scenarios, generate initial trust scores, and use the initial trust scores as the initial attributes of dynamic trust graph nodes to construct dynamic trust graphs;

[0160] Zero-knowledge proof credential module: Receives access requests initiated by data requesters and generates zero-knowledge proof credentials based on the zero-knowledge proof protocol and local sensitive context information;

[0161] Dynamic Trust Assessment Module: Monitors order status changes, logistics node updates, and abnormal alarm event signals in the supply chain business flow in real time, inputs the event signals as weighting factors into the dynamic trust graph, calculates and outputs the real-time updated dynamic trust score;

[0162] Access control policy matching module: Based on the numerical range of the dynamic trust score, it matches the corresponding data desensitization algorithm strength level and field visibility range policy to generate a fine-grained access control parameter group containing masking rules or aggregation granularity.

[0163] Differentiated data delivery module: Based on the fine-grained access control parameter group, the target supply chain data is processed differently. If the dynamic trust score is higher than the preset high trust threshold, the complete plaintext data is output. If it is lower than the preset low trust threshold, the trend data is output after key value masking or aggregation.

[0164] Data storage module: Hash the access request details, zero-knowledge proof verification results, dynamic trust score change trajectory and final data delivery form to generate tamper-proof audit storage data and write it into the blockchain ledger;

[0165] Risk handling module: Monitors unauthorized access patterns in audit and evidence storage data. If potential risky behavior is detected, it triggers a smart contract to execute a trust score freeze command.

[0166] In one embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a code vulnerability analysis and discovery method for network security software.

[0167] In one embodiment, a computer device is provided, the computer device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of any of the methods in the above method embodiments.

[0168] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of any of the methods described in the above method embodiments.

[0169] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Claims

1. Supply chain data sharing methods for cross-enterprise collaboration, specifically including: S1: Obtain enterprise qualification data, historical performance records and third-party credit scores of each participant in the cross-enterprise collaboration scenario, generate an initial trust score, and use the initial trust score as the initial attribute of the dynamic trust graph node to construct the dynamic trust graph; S2: Receive the access request initiated by the data requester, and generate a zero-knowledge proof credential based on the zero-knowledge proof protocol using local sensitive context information; S3: Monitor order status changes, logistics node updates and abnormal alarm event signals in the supply chain business flow in real time, input the event signals as weighting factors into the dynamic trust graph, calculate and output the real-time updated dynamic trust score; S4: Based on the numerical range of the dynamic trust score, match the corresponding data desensitization algorithm strength level and field visibility range strategy to generate a fine-grained access control parameter group containing masking rules or aggregation granularity. S5: Based on the fine-grained access control parameter group, perform differentiated processing on the target supply chain data. If the dynamic trust score is higher than the preset high trust threshold, output complete plaintext data; if it is lower than the preset low trust threshold, output trend data that has been processed by key value masking or aggregation. S6: Perform hash operations on the access request details, zero-knowledge proof verification results, dynamic trust score change trajectory, and final data delivery form to generate tamper-proof audit evidence data and write it into the blockchain ledger.

2. The supply chain data sharing method for cross-enterprise collaboration according to claim 1, characterized in that, Following S6, the following is also included: S7: Monitor unauthorized access patterns in audit and evidence storage data. If potential risky behavior is detected, trigger the smart contract to execute a trust score freeze instruction.

3. The supply chain data sharing method for cross-enterprise collaboration according to claim 1, characterized in that, In S2, zero-knowledge proof credentials are generated using local sensitive context information based on a zero-knowledge proof protocol to verify the compliance of the operating environment without disclosing the specific geographical location and device fingerprint.

4. The supply chain data sharing method for cross-enterprise collaboration according to claim 1, characterized in that, In S1, a third-party credit score is obtained based on a third-party credit scoring report.

5. The supply chain data sharing method for cross-enterprise collaboration according to claim 1, characterized in that, The historical performance record data in S1 includes transaction timestamps, order amounts, delivery delay durations, and default flags.

6. The supply chain data sharing method for cross-enterprise collaboration according to claim 1, characterized in that, Step S3 specifically includes: The order status change logs, logistics node update messages and abnormal alarm signals in the supply chain business flow are parsed and standardized to eliminate data format heterogeneity and extract standard business event sequences containing timestamps, event types and associated entity identifiers. Based on the standard business event sequence, semantic matching and risk level labeling are performed using a predefined risk assessment mapping rule base to transform discrete event types into initial event weight vectors with positive and negative directionality and intensity quantification. The initial event weight vector is weighted and fused by element-wise multiplication with a time decay coefficient and a behavior frequency statistical factor to generate a dynamic event influence coefficient that can reflect recent behavior trends and suppress historical noise interference. Based on the dynamic event impact coefficient, the corresponding enterprise user node and its adjacent edges are located in the dynamic trust graph. The node attribute iterative update process under the graph propagation algorithm is executed to calculate the real-time dynamic trust score that integrates the global topology and local behavior feedback.

7. The supply chain data sharing method for cross-enterprise collaboration according to claim 1, characterized in that, S4 specifically includes: Obtain the real-time updated dynamic trust score output by the dynamic trust graph, and standardize the dynamic trust score to eliminate the influence of different units and generate a standardized trust metric. Receive the standardized trust metric and map the standardized trust metric to a predefined trust level sequence to determine the dynamic trust level label to which the current requester belongs. Read the dynamic trust level label, call the pre-built de-identification algorithm strength level configuration table, and filter out the target de-identification algorithm strength level corresponding to the dynamic trust level label; Load the target desensitization algorithm strength level, combine it with supply chain data field sensitivity metadata, and generate a masking rule set and aggregation granularity for specific data fields; By integrating the masking rule set with the aggregation granularity, a structured fine-grained access control parameter group is constructed based on the parameter encapsulation protocol, which serves as the direct execution basis for subsequent data differentiation processing.

8. A supply chain data sharing system for cross-enterprise collaboration, characterized in that: The system includes: Dynamic Trust Graph Module: Acquire enterprise qualification data, historical performance records and third-party credit scores of each participant in cross-enterprise collaboration scenarios, generate initial trust scores, and use the initial trust scores as the initial attributes of dynamic trust graph nodes to construct dynamic trust graphs; Zero-knowledge proof credential module: Receives access requests initiated by data requesters and generates zero-knowledge proof credentials based on the zero-knowledge proof protocol and local sensitive context information; Dynamic Trust Assessment Module: Monitors order status changes, logistics node updates, and abnormal alarm event signals in the supply chain business flow in real time, inputs the event signals as weighting factors into the dynamic trust graph, calculates and outputs the real-time updated dynamic trust score; Access control policy matching module: Based on the numerical range of the dynamic trust score, it matches the corresponding data desensitization algorithm strength level and field visibility range policy to generate a fine-grained access control parameter group containing masking rules or aggregation granularity. Differentiated data delivery module: Based on the fine-grained access control parameter group, the target supply chain data is processed differently. If the dynamic trust score is higher than the preset high trust threshold, the complete plaintext data is output. If it is lower than the preset low trust threshold, the trend data is output after key value masking or aggregation. Data storage module: Hash the access request details, zero-knowledge proof verification results, dynamic trust score change trajectory and final data delivery form to generate tamper-proof audit storage data and write it into the blockchain ledger; Risk handling module: Monitors unauthorized access patterns in audit and evidence storage data. If potential risky behavior is detected, it triggers a smart contract to execute a trust score freeze instruction.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that... When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1-7.

Images