Cloud architecture-oriented multi-dimensional credit data processing system

Abstract

The invention relates to the technical field of big data processing and information security, and discloses a cloud architecture-oriented multi-dimensional credit data processing system. Comprising a multi-source data acquisition and standardization module, a multi-dimensional space-time topology mapping engine, a query intention analysis and topology preheating module, a cold and hot layered encryption storage scheduler, a path integrity verification and cache management module and an event driving and incremental calculation module. According to the method, a multi-dimensional spatio-temporal topological graph containing entities and field domain nodes is constructed, and the influence of spatial attributes on credit is quantified; performing cold and hot layered encryption storage by utilizing query mode recognition; constructing a cache fingerprint based on the encrypted object to verify the path integrity; and in response to data change, performing dynamic incremental calculation by using a credit potential energy conduction model in combination with a consistency verification result. According to the method, on the premise of ensuring data privacy and consistency, the calculation efficiency and evaluation accuracy of the large-scale credit graph are improved.

Description

Technical Field

[0001] This invention relates to the fields of big data processing and information security technology, specifically a multi-dimensional credit data processing system oriented towards cloud architecture. Background Technology

[0002] With the rapid development of the digital economy, the data sources for credit reporting systems have expanded from traditional credit records to multi-source heterogeneous data from taxation, business registration, and judicial sources. To cope with the storage and computing needs of massive amounts of data, credit reporting systems are gradually migrating to cloud-native architectures, and building large-scale enterprise relationship graphs using the elastic resources of cloud platforms has become a mainstream industry trend.

[0003] Existing credit data processing technologies based on credit graphs still have limitations in terms of model dimensions, computational timeliness, and data security. In terms of model dimensions, traditional credit graph construction mainly relies on explicit business connections such as equity investments, guarantee chains, or supply chain transactions. While this modeling approach can track the risk of fund flows, it often overlooks the implicit impact of the geographical environment on the operational stability of the credit subject. For example, policy changes in industrial parks and fluctuations in the regional economic climate are often not effectively represented in the graph, making it difficult for risk assessment models to identify systemic risks caused by regional factors.

[0004] Regarding computational timeliness and data consistency, with the exponential growth of graph size, credit transmission calculations based on full graph traversal are often time-consuming, making it difficult to meet the needs of real-time credit inquiries. Although existing technologies generally introduce caching mechanisms to improve response speed, they lack efficient verification methods for complex topologies when facing high-frequency concurrent data changes. When there are minor changes in the underlying node attributes or relationships, the system often struggles to accurately determine whether the cached path calculation results are still valid, easily leading to delayed credit scoring or outputting incorrect evaluation results based on dirty data.

[0005] In cloud architecture, existing technologies often face a trade-off between performance and security when it comes to data storage and privacy protection. To ensure the security of sensitive credit data, strong encryption storage of the entire dataset significantly increases I / O overhead, leading to query latency. Conversely, reducing encryption strength to pursue faster retrieval exposes data leakage risks in shared resource pools such as object storage. Existing storage scheduling solutions lack the ability to dynamically detect the access frequency of graph data, making it difficult to achieve low-cost, high-strength privacy protection for cold data while simultaneously meeting high-concurrency, low-latency query requirements. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a multi-dimensional credit data processing system for cloud architecture, which solves the problems of existing cloud-based credit systems lacking consideration of the geospatial environment dimension and struggling to balance computational real-time performance, data consistency verification efficiency, and end-to-end privacy and security when processing massive amounts of map data.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a multi-dimensional credit data processing system for cloud architecture, comprising:

[0008] The multi-source data acquisition and standardization module is used to map the acquired raw data into standardized feature vectors and use change data capture technology to monitor changes in the standardized feature vectors to generate data change events.

[0009] A multi-dimensional spatiotemporal topology mapping engine, connected to the multi-source data acquisition and standardization module, is used to receive the standardized feature vector, construct a multi-dimensional spatiotemporal topology map containing entity nodes and field nodes, and calculate the weights of the associated edges and belonging edges between nodes.

[0010] The query intent analysis and topology warm-up module is used to analyze the query log stream to identify the target query pattern and generate a preloading instruction for the corresponding subgraph range based on the target query pattern.

[0011] The hot and cold tiered encrypted storage scheduler module is connected to the query intent analysis and topology preheating module. It is used to schedule node data between the hot zone of the memory database and the cold zone of the object storage according to the preloading instruction, and to encrypt the node data that sinks to the cold zone of the object storage to generate encrypted data objects.

[0012] The path integrity verification and cache management module is used to construct a cache fingerprint based on the encrypted data objects corresponding to each node on the target transmission path in the multidimensional spatiotemporal topology map, and before the credit potential calculation is initiated, read the encrypted data objects of each node on the target transmission path to generate a real-time fingerprint, and compare the real-time fingerprint with the cache fingerprint to generate a consistency verification result.

[0013] The event-driven and incremental calculation module is connected to the multi-source data acquisition and standardization module and the path integrity verification and cache management module, respectively. It is used to respond to the data change event, dynamically calculate the credit potential based on the credit potential transmission model and the consistency verification result, obtain the potential change of the affected node, and update the credit score according to the potential change of the affected node.

[0014] Furthermore, the system also includes an application service interface module, connected to the event-driven and incremental calculation module, used to map the updated credit score to a credit rating.

[0015] Furthermore, the multi-source data acquisition and standardization module is used to access external heterogeneous data sources to collect raw data associated with multidimensional credit data.

[0016] Furthermore, asymmetric encryption is performed on the node data that has sunk to the object storage cold zone to generate encrypted data objects.

[0017] Furthermore, the credit rating is a discrete credit rating.

[0018] Furthermore, the path integrity verification and cache management module is used to construct a topological hash fingerprint chain based on the encrypted data objects corresponding to each node on the target transmission path in the multidimensional spatiotemporal topology map, and store it as a cache fingerprint in the hot zone of the memory database. Before the credit potential calculation is initiated, the encrypted data objects of each node on the target transmission path are reread to generate a real-time fingerprint, and the consistency between the real-time fingerprint and the cache fingerprint is compared to generate a consistency verification result.

[0019] Furthermore, the application service interface module is also used to generate and deliver encrypted credit reports after obtaining discrete credit ratings.

[0020] Preferably, the multi-source data acquisition and standardization module is specifically used for:

[0021] The original data is converted into a standardized feature vector through a preset standardization function. For continuous numerical fields, logarithmic transformation is used for normalization, and for discrete fields, numerical encoding is used.

[0022] The data change event includes the credit subject's unique identifier, the amount of feature change, the timestamp, and the operation type;

[0023] The multi-source data acquisition and standardization module generates and pushes the data change event only when the calculated feature change is greater than zero.

[0024] Preferably, the multidimensional spatiotemporal topology mapping engine constructs the field nodes in the following ways:

[0025] Parse regional keywords with administrative or economic functions from the address field of the credit subject, and materialize the regional keywords into independent nodes in the graph;

[0026] The multidimensional spatiotemporal topology mapping engine establishes the belonging edges of entity nodes to the field nodes; the formula logic for calculating the weights by the multidimensional spatiotemporal topology mapping engine is as follows:

[0027] The weight is equal to the product of the global normalization coefficient, the edge type-based strength factor, and the time decay factor;

[0028] The time decay factor is calculated using an exponential function with the natural constant as the base and the negative of the product of the time decay constant and the time difference as the exponent.

[0029] Preferably, the query intent analysis and topology warm-up module calculates the frequency of query patterns using a sliding time window statistical model.

[0030] Preferably, the calculation logic for the frequency of the target query pattern is as follows: within the sliding window, the weighted sum of the requests matching the target query pattern is divided by the window time length;

[0031] When the frequency of the calculated query pattern exceeds a preset popularity threshold, the preloading instruction is generated.

[0032] The preload instruction includes a list of unique identifiers for the nodes to be loaded and the target storage level.

[0033] Preferably, the logic configuration for the cold / hot tiered encrypted storage scheduler module to perform data encryption and decryption is as follows:

[0034] When the node data needs to be persisted to the object storage cold area, the serialized node data is encrypted using the platform public key preset by the system to generate the encrypted data object;

[0035] When the node data needs to be loaded into the hot zone of the memory database, the encrypted data object is decrypted and restored using the platform private key that is only stored in the memory isolation zone.

[0036] Preferably, the path integrity verification and cache management module constructs the topological hash fingerprint chain in the following ways:

[0037] Obtain the encrypted data object corresponding to each node on the target transmission path in the object storage cold zone;

[0038] The encrypted data object is directly subjected to hash digest calculation to generate a node-level fingerprint;

[0039] According to the topological order of the target propagation path, the node-level fingerprints of all nodes on the target propagation path are concatenated and spliced.

[0040] A hash operation is performed again on the concatenated data to generate a unique fingerprint chain signature for the target transmission path.

[0041] Preferably, the path integrity verification and cache management module pre-calculates the composite transmission coefficient of the target transmission path, and the composite transmission coefficient is equal to the product of the weights of all edges on the target transmission path;

[0042] The credit potential energy calculation includes: selecting to call the composite conduction coefficient for fast calculation or performing a jump-by-hop traversal calculation to obtain the potential energy change of the affected node;

[0043] The fast calculation logic executed by the event-driven and incremental calculation module is as follows:

[0044] When the path integrity verification and cache management module confirms that the real-time fingerprint is consistent with the cache fingerprint, it directly calls the product of the composite transmission coefficient and the initial credit potential as the potential increment of the target node.

[0045] Preferably, the credit potential transmission model defined in the event-driven and incremental calculation module includes: multiplying the feature change amount in the data change event with the feature sensitivity matrix to obtain the initial credit potential impact;

[0046] The initial credit potential is injected into the multidimensional spatiotemporal topology graph as the output energy of the source node. For any affected node in the multidimensional spatiotemporal topology graph, its accumulated credit potential is calculated. The accumulated credit potential is equal to the sum of the products of the output energy of all incoming neighbor nodes and the corresponding edge weights.

[0047] A dynamic sensitivity threshold is set. When the accumulated credit potential exceeds the dynamic sensitivity threshold, the node's credit score is updated and the energy is transmitted to downstream nodes.

[0048] Preferably, the application service interface module has a built-in dynamic quantile mapping algorithm, which is used to calculate the values ​​of different percentiles as boundary thresholds for grade division based on the score distribution of the full range of credit subjects, and to determine multiple threshold intervals based on multiple boundary thresholds; and to determine the credit level of the target credit subject based on the threshold interval into which the credit score falls.

[0049] Preferably, the hop-by-hop traversal calculation logic executed by the event-driven and incremental calculation module is as follows:

[0050] When the consistency verification result is a verification failure, the rollback calculation mode is initiated;

[0051] Starting from the node associated with the data change event, the credit potential energy is calculated and accumulated hop by hop along the edge structure of the multidimensional spatiotemporal topology graph using a breadth-first search or depth-first search algorithm to obtain the potential energy change of the target node.

[0052] This invention provides a multi-dimensional credit data processing system for cloud architecture, which has the following advantages:

[0053] This invention constructs a multi-dimensional spatiotemporal topological map containing entity nodes and field nodes, materializing the geospatial attributes of credit subjects as functional nodes in the map and establishing attribution associations between entities and fields. This design overcomes the limitations of traditional credit models that only focus on commercial relationships such as equity or transactions, effectively quantifying the radiating impact of environmental factors such as park policies and geographical location on corporate credit. This allows for a more accurate identification of systemic risks caused by regional economic fluctuations or industrial cluster effects, improving the comprehensiveness and accuracy of risk assessment dimensions.

[0054] This invention utilizes a topological hash fingerprint chain built upon underlying encrypted objects. This system can quickly verify the validity of the cached path before computation is initiated. When it is confirmed that the data has not been tampered with and the topological structure is stable, the system directly calls the composite transmission coefficient of the cache to perform a jump calculation with constant complexity. This avoids the computational power consumption caused by repeatedly traversing the entire graph. While ensuring that the credit score can respond to data changes in real time, cryptographic means are used to prevent computational deviations caused by cache pollution or data inconsistency.

[0055] This invention employs a hot-and-cold tiered encrypted storage scheduling strategy, optimizing the storage cost of massive graph data while ensuring data physical security. Based on query intent analysis, the system automatically schedules frequently accessed node data to a high-performance memory hot zone, while relegating low-frequency data to a low-cost object storage cold zone and performing asymmetric encryption. The encrypted data objects not only serve as static storage entities but also directly as the trust foundation for consistency verification. Combined with digital envelope technology at the application service interface, this achieves end-to-end data privacy protection from underlying storage and computation to final delivery. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the logical structure of a cloud-oriented multi-dimensional credit data processing system according to an embodiment of the present invention.

[0057] Figure 2 This is a schematic diagram illustrating the principle of topological hash fingerprint chain construction and integrity verification in an embodiment of the present invention;

[0058] Figure 3 This is a flowchart illustrating the event-driven and incremental calculation logic of an embodiment of the present invention.

[0059] Figure 4 This is a schematic diagram of the logical structure of the multi-source data acquisition and standardization module in an embodiment of the present invention;

[0060] Figure 5 This is a schematic diagram of the logical structure of the multidimensional spatiotemporal topology mapping engine according to an embodiment of the present invention;

[0061] Figure 6This is a schematic diagram of the logical structure of the path integrity verification and cache management module in an embodiment of the present invention;

[0062] Figure 7 This is a schematic diagram of the logical structure of the event-driven and incremental calculation module in an embodiment of the present invention;

[0063] Figure 8 This is a schematic diagram of the logical structure of the application service interface module in an embodiment of the present invention. Detailed Implementation

[0064] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Please see Figures 1 to 3 This invention provides a multi-dimensional credit data processing system for cloud architecture, comprising: a multi-source data acquisition and standardization module, a multi-dimensional spatiotemporal topology mapping engine, a query intent analysis and topology preheating module, a cold and hot hierarchical encrypted storage scheduler module, a path integrity verification and cache management module, an event-driven and incremental calculation module, and an application service interface module.

[0066] like Figure 4 As shown, the multi-source data acquisition and standardization module is used to access external heterogeneous data sources, convert the acquired raw data into standardized feature vectors, and use a change data capture method to monitor changes in the standardized feature vectors and generate data change events containing the changes.

[0067] Specifically, the multi-source data acquisition and standardization module in this embodiment includes an interface adaptation unit, a data cleaning and vectorization unit, and a change capture unit.

[0068] The interface adaptation unit is configured as the system's communication front-end, used to establish physical connections and protocol handshakes with external heterogeneous data sources. These external heterogeneous data sources are logically divided into government data sources and commercial data sources. For government data sources, the interface adaptation unit accesses the tax management system, judicial case management system, social insurance management system, and power supply system via dedicated lines or encrypted channels to obtain government-related data, including but not limited to enterprise tax ratings, administrative penalty records, pending litigation information, number of social security contributors, and annual electricity consumption. For commercial data sources, the interface adaptation unit accesses third-party credit reporting agencies or supply chain finance platforms via API interfaces to obtain commercial-related data such as enterprise registration information, equity structure, upstream and downstream supply chain counterparties, and transaction amounts. Given the potential differences in data interaction protocols used by different data sources, the interface adaptation unit integrates a multi-protocol parser to uniformly convert the received raw data stream into a system-defined JSON or XML intermediate format, forming the raw dataset to be processed.

[0069] The data cleaning and vectorization unit is connected to the interface adaptation unit and is used to perform preprocessing and feature mapping on the original dataset. The preprocessing includes filling missing fields with default values, truncating outlier values, and deduplicating duplicate records. Based on this, the data cleaning and vectorization unit performs standardization mapping logic, that is, through a preset standardization function. any original data Transform into a standardized feature vector with uniform dimensions. .

[0070] ;

[0071] In the formula, This represents the standardized feature vector. Represents the normalized function. Indicates the first Standardized values ​​of each feature Represents the total dimension of the features. This represents the original feature data or the original dataset. This indicates the transpose operation.

[0072] For continuous numerical fields, max-min normalization or logarithmic transformation is used to map them to the [0,1] interval to eliminate the influence of dimensions. The normalization formula is as follows:

[0073] ;

[0074] In the formula, This represents the standardized numerical value. Represents the original data value. This represents the minimum value of this feature in historical data. This represents the maximum value of this feature in historical data. Represents the natural logarithm operation.

[0075] For discrete enumerated fields, numeric encoding or one-hot encoding is used. For example, tax ratings can be mapped to discrete numerical values: For binary state fields, such as whether there is pending litigation, the value is mapped to 1.0 if it exists, and 0.0 otherwise. Through the above processing, the heterogeneous raw data is transformed into standardized feature vectors suitable for subsequent graph computations and matrix operations.

[0076] The change capture unit is used to monitor standardized feature vectors. The state changes are the foundation for implementing an event-driven mechanism. To reduce intrusion into the source business system, the change capture unit employs log parsing-based change data capture technology. Specifically, the change capture unit simulates a database slave node protocol, monitoring and parsing the binary logs or system operation logs of the source database in real time.

[0077] When the system detects that the feature data of a credit subject has been inserted, updated, or deleted, the change capture unit obtains a snapshot of the data before the change. Compared with the changed data snapshot And calculate the characteristic change. .like If a substantial change is detected, the change capture unit immediately generates a data change event. The data structure definition for the data change event is as follows:

[0078] ;

[0079] In the formula, Indicates a data change event. A unique identifier representing the subject of credit reporting. Indicates the characteristic change. Represents a timestamp. Indicates the operation type.

[0080] generated The data is directly pushed to a high-throughput message queue within the system for consumption by subsequent modules, thus decoupling data collection from subsequent risk calculations.

[0081] like Figure 5 As shown, the multi-dimensional spatiotemporal topology mapping engine is connected to the multi-source data acquisition and standardization module to receive the standardized feature vector and construct a multi-dimensional spatiotemporal topology map in the cloud; the multi-dimensional spatiotemporal topology map includes entity nodes representing credit subjects, field nodes representing geographical functional areas, and associated edges connecting each node.

[0082] Specifically, the multidimensional spatiotemporal topology mapping engine in this embodiment includes a node instantiation unit, a relationship construction unit, and a weight dynamic calculation unit. The multidimensional spatiotemporal topology mapping engine runs in a cloud-based graph database environment and is used to map the flattened feature vectors output by the multi-source data acquisition and standardization module into a multidimensional heterogeneous graph with a topological structure.

[0083] The node instantiation unit is used to construct the node set of the graph. Unlike traditional social networks that focus only on homogeneous nodes, the node set in this embodiment of the invention consists of two mutually exclusive subsets: the entity node set and the entity node set. With field node set .

[0084] For the set of entity nodes The node instantiation unit creates a graph node for each credit reporting entity with a unique identifier. Each entity node... The internal storage contains the standardized feature vector of the credit subject. This serves as the baseline state for subsequent risk calculations.

[0085] For the set of field nodes The node instantiation unit performs address semantic parsing and spatial clustering operations. It extracts regional keywords with administrative or economic functions from the enterprise's registered address or business address field and materializes these regions into independent nodes in the graph, i.e., field nodes. Field nodes not only serve as structural supports for the graph, but also as centers for the convergence and dissemination of regional credit risks.

[0086] The relation building unit is used to construct the edge set of the graph. This clarifies the information transmission paths between nodes. The edge set contains the set of associated edges. With the set of belonging edges .

[0087] Related edge set Defined between two entity nodes, i.e. The relationship building unit establishes directed edges between the relevant entity nodes based on equity penetration chains, supply chain transaction records, and guarantee contracts in the business data.

[0088] Set of belonging edges Defined between entity nodes and field nodes, i.e. The relationship building unit establishes belonging edges from entity nodes to field nodes based on the spatial attributes of entities. These belonging edges represent the degree to which an entity is affected by its local environment, allowing risks to be transmitted along paths between enterprises, parks, and other enterprises, thereby simulating the risk resonance effect of enterprises within the same park in the real world due to policy commonalities or geographical proximity.

[0089] The weight dynamic calculation unit is used to quantize each edge. The risk transmission capability, i.e., the calculation weight. Considering the closeness and timeliness of business relationships, the weighted dynamic calculation unit adopts a comprehensive calculation model that includes relationship type strength and time decay factor.

[0090] The specific weight calculation formula is defined as follows:

[0091] ;

[0092] In the formula, It is a global normalization coefficient used to constrain the calculation results within the numerical range allowed by the system, preventing numerical explosion in multi-hop propagation;

[0093] For edge type-based strength factor functions, for associated edges If the relationship type is equity investment, The value is taken as the shareholding ratio; if it is a guarantee relationship, The value is the ratio of the guarantee amount to the net assets of the guaranteed party.

[0094] For the belonging edge , The value is a preset field influence coefficient, which is determined by the administrative level of the field;

[0095] , indicating the current time With the time of relationship establishment The time difference between them;

[0096] The time decay constant is obtained through the exponential decay function. It can automatically reduce the risk transmission weight of old relationships, thereby reflecting the timeliness of credit risk.

[0097] When the multi-source data acquisition and standardization module captures the establishment of new relationships or the termination of old relationships through the CDC (Change Data Capture) mechanism, the relationship construction unit adds or deletes edges in the graph in real time, and the weight dynamic calculation unit updates the weight values ​​of the relevant edges synchronously to ensure that the multi-dimensional spatiotemporal topology graph always reflects the current business and spatial relationship status.

[0098] The query intent analysis and topology warm-up module is used to analyze the system's query logs to identify target query patterns and generate preloading instructions based on the subgraph range corresponding to the target query pattern.

[0099] Specifically, the query intent analysis and topology warm-up module in this embodiment includes a log stream analysis unit and a warm-up instruction generation unit. This module establishes a feedback loop between the business query layer and the underlying storage layer, utilizing the principles of temporal and spatial locality of access to address the high cold-start latency issue of large-scale graph data in cloud environments.

[0100] The log stream analysis unit is used to collect and parse query request logs from external users in real time. The log stream analysis unit does not focus on the discrete state of a single query, but rather on aggregated query behavior within a time window. This unit maintains a statistical model based on a sliding time window to extract four-tuple feature vectors from the log stream. ,in The identifier of the user organization that initiated the query. For the target credit reporting entity to be queried, To query the associated attributes of the target (such as industry, park location), For timestamps.

[0101] To accurately identify target query patterns that should be pre-warmed, the log stream analysis unit employs a streaming frequent itemset mining algorithm. The system has a pre-defined set of query patterns. This includes business models such as bulk screening of specific industrial parks by specific financial institutions and upstream and downstream penetration inquiries of core enterprises in specific supply chains. The log stream analysis unit continuously calculates the frequency of occurrence of each model within the current sliding window. .

[0102] ;

[0103] In the formula, This indicates the frequency calculation result. Indicates the duration of the sliding time window. This represents the total number of requests within the sliding time window. Indicates the first This request, Indicates the target query pattern or condition. Indicates an indicator function, Represents the weighting function. Indicates the first The time of this request.

[0104] When the calculated frequency Exceeding the system's preset heat threshold When the log stream analysis unit identifies the pattern as the target query pattern, marks it as active, and sends the corresponding pattern parameters to the preheating instruction generation unit.

[0105] The preheating instruction generation unit is used to transform the abstract business query mode into a specific graph subgraph preloading instruction. It internally stores the graph's metadata index and can quickly locate the node range involved in the multidimensional spatiotemporal topology graph based on the mode parameters.

[0106] Specifically, if the received pattern is a regional risk assessment, the parameter is a field node. The preheating instruction generation unit retrieves all nodes in the field that have established belonging edges with the field node through an index. The set of entity nodes If the received pattern is supply chain penetration, the parameter is the core enterprise node. Then the search is based on Starting from The set of skip neighbor nodes.

[0107] After determining the sub-graph range to be preheated, the preheating instruction generation unit generates standardized preloading instructions. It contains a list of unique identifiers for nodes to be loaded and the target storage level. The subgraph range is a local substructure composed of the set of nodes and edge sets that are strongly associated with the target query pattern in the multidimensional spatiotemporal topology graph. It is the core scheduling object of the preloading instruction, and its specific range is dynamically determined by the query pattern parameters.

[0108] The preload directive formatting definition is:

[0109] ;

[0110] In the formula, This indicates a preload instruction. A list of unique identifiers representing the nodes to be loaded. Indicates the target storage level. Indicates priority.

[0111] Generated preload instructions The data is sent directly to the cold / hot tiered encrypted storage scheduler module, triggering a data migration operation on the underlying storage medium. Through this mechanism, when subsequent large-scale batch query or computation requests actually arrive, the associated graph data has already been pre-residing in high-performance memory, thereby eliminating the I / O wait time caused by loading and decrypting data from the object storage cold zone and realizing dynamic adaptation of storage resources to business intent.

[0112] The hot and cold tiered encrypted storage scheduler module is connected to the query intent analysis and topology preheating module. It is used to schedule the node data in the multidimensional spatiotemporal topology map between the hot zone of the memory database and the cold zone of the object storage according to the preloading instruction and data activity, and to perform asymmetric encryption on the node data stored in the cold zone of the object storage to generate encrypted data objects.

[0113] Specifically, the cold / hot tiered encrypted storage scheduler module in this embodiment includes a tiered management unit and a data encryption / decryption unit. As the system's data foundation, it is responsible for striking a balance between computational efficiency and storage cost, and ensuring the physical security of stationary data.

[0114] The hierarchical management unit manages the physical storage location of multidimensional spatiotemporal topology map data, constructing a two-tiered storage architecture consisting of a hot zone in memory database and a cold zone in object storage. The hot zone in memory database employs a memory-based key-value storage cluster or in-memory graph database, providing microsecond-level random read / write capabilities for storing frequently accessed active node data and pre-warming data. The cold zone in object storage uses distributed block storage or object storage services for persistent storage of the full map data and historical snapshots; it is inexpensive but has relatively high access latency.

[0115] The hierarchical management unit has a built-in state machine that maintains an activity state flag for each node in the graph. Based on preloading instructions from the query intent analysis and topology warm-up modules, and the real-time access frequency of data, a dynamic data migration strategy is executed. When a preloading instruction for a specific subgraph is received, or when a cold node is detected to be accessed multiple times within a short period of time, the hierarchical management unit marks that node as... This triggers a float operation from the cold zone to the hot zone. Conversely, for hot zones that have not been accessed for a long time (exceeding a preset time threshold),... The hierarchical management unit marks the node as... This triggers a sinking operation from the hot zone to the cold zone, releasing scarce memory resources.

[0116] The data encryption / decryption unit is deployed at the I / O boundary between the cold and hot zones to perform secure data conversion. To meet the high-security requirements for credit data protection and to provide a cryptographic basis for subsequent path integrity verification, the data encryption / decryption unit uses the national standard SM2 asymmetric encryption algorithm to encrypt and encapsulate the data that has settled into the cold zone.

[0117] Specifically, when node data When data needs to be persisted to the object storage cold storage, the data encryption / decryption unit first serializes the standardized feature vectors, associated edge lists, and attribute information contained in the node data into a binary stream. Subsequently, the system's pre-set platform public key was used. Perform encryption operations on the binary stream to generate an encrypted data object. The encryption process can be formally described as follows:

[0118] ;

[0119] In the formula, This represents the encrypted object. This indicates that the SM2 encryption algorithm is used to encrypt the data. This indicates the node data to be encrypted. This represents the public key.

[0120] The encrypted data object It is not only the storage format of the data, but also the physical entity used for hash fingerprint calculation to verify path integrity. Any verification of data tampering is based on this encrypted object itself, thus ensuring that the verification process does not depend on decryption operations, guaranteeing the atomicity and security of the verification.

[0121] When the hierarchical management unit performs the up-float operation, the data encryption / decryption unit reads from the cold zone. Using the platform's private key, which is stored only in a memory-isolated area. Decrypt and restore:

[0122] ;

[0123] In the formula, This represents the decrypted data. This indicates that the SM2 decryption algorithm is used to decrypt the data. Represents the encrypted data object. This represents the private key.

[0124] The decrypted node data is loaded into the hotspot of the in-memory database.

[0125] like Figure 2 and Figure 6 As shown, the path integrity verification and cache management module is connected to the cold and hot hierarchical encrypted storage scheduler module. It is used to construct a topological hash fingerprint chain based on the encrypted data objects corresponding to each node on the target transmission path in the multidimensional spatiotemporal topology map and store it as a cache fingerprint in the hot zone of the memory database. Before the credit potential calculation is initiated, it rereads the encrypted data objects of each node on the target transmission path to generate a real-time fingerprint and compares the consistency between the real-time fingerprint and the cache fingerprint to generate a consistency verification result.

[0126] Specifically, the path integrity verification and cache management module in this embodiment includes a path analysis and cache construction unit, a fingerprint chain generation unit, and a consistency verification execution unit. Its purpose is to establish a cross-level trust mechanism to ensure that cached data used for fast computation strictly corresponds to tamper-proof encrypted entities in the underlying physical storage.

[0127] The path analysis and cache construction unit is used to identify target transmission paths in a multidimensional spatiotemporal topology map and pre-calculate the composite attributes of the paths. In complex credit transmission networks, specific supply chain links or regional linkage structures often bear most of the risk traffic. The path analysis and cache construction unit continuously counts the frequency of transmission requests passing through each node, and when a certain path... When the access popularity exceeds a preset threshold, it is marked as a target propagation path.

[0128] For the target propagation path, the path analysis and cache construction unit pre-extracts all edges on the target propagation path. weight And calculate the composite conductivity coefficient. This composite transmission coefficient represents the risk energy originating from the path's starting point. Conducted to the endpoint The total attenuation rate is calculated using the following formula:

[0129] ;

[0130] In the formula, Represents the composite conductivity coefficient. Indicates the number of nodes on the target propagation path. Indicates the first step on the target transmission path 1 node Indicates the path from the node To the node The weight or transmission coefficient.

[0131] The fingerprint chain generation unit is used to bind a topological hash fingerprint chain with cryptographic strength to each cache composite conduction coefficient in order to solve the cache consistency problem.

[0132] Specifically, the fingerprint chain generation unit requests each node on the target propagation path from the storage scheduler. Corresponding encrypted data object The fingerprint chain generation unit does not request the decrypted plaintext, but directly performs hash digest calculations on the encrypted object itself to generate node-level fingerprints.

[0133] ;

[0134] In the formula, Represents a node hash value, Represents a hash function. Represents the encrypted data object. Indicates the first step on the target transmission path Each node.

[0135] Based on this, the fingerprint chain generation unit concatenates the node-level fingerprints of all nodes along the target transmission path according to the topological order of the path, and performs hash operations again to generate a unique fingerprint chain signature for the target transmission path. .

[0136] ;

[0137] In the formula, A signature representing the target propagation path. Represents a node hash value, Indicates the concatenation operator. This represents a hash function.

[0138] generated The fingerprint is marked as a cache fingerprint, a key-value pair mapping relationship is established with the composite conduction coefficient, and they are persistently stored together in the hot zone of the memory database.

[0139] The consistency verification execution unit is deployed at the entry point of the computation request and is used to perform lightweight integrity verification before the event-driven and incremental computation modules initiate the actual credit potential calculation. This is done when the computation module requests to use the composite transmission coefficient of a certain path. At that time, the consistency verification execution unit triggers the real-time verification process.

[0140] During the verification process, the consistency verification execution unit accesses the object storage cold zone, rereads the latest encrypted data objects corresponding to each node on the current target propagation path, and calculates the real-time fingerprint on-site using the same hash algorithm and cascading rules as the fingerprint chain generation unit. Subsequently, Compared with the original signature stored in the cache Perform the comparison to generate the consistency verification result:

[0141] like If the consistency verification result is "verification passed," it indicates that the underlying data of all nodes on the path has not changed since the cache was established, and the storage medium has not been tampered with. The composite conduction coefficient of the cache is [value missing]. It is safe and effective; the consistency verification execution unit returns a verification pass signal to the computing module. This allows the calculation module to perform skip calculations directly.

[0142] like If the consistency verification result is "verification failed," it indicates that the data at a certain node on the path has changed. The consistency verification execution unit determines that the cache is invalid and returns a verification failure signal to the calculation module, forcibly triggering a hop-by-hop calculation process, and simultaneously notifying the path analysis and cache construction unit to update the cache entries for that path. Through this pre-computation verification mechanism, the system can enjoy the performance benefits brought by caching while using cryptographic means to eliminate the risk transmission error caused by data inconsistency.

[0143] The consistency verification result is sent to the event-driven and incremental calculation module as a control signal for selecting a calculation strategy.

[0144] like Figure 3 and Figure 7 As shown, the event-driven and incremental calculation module is connected to the multi-source data acquisition and standardization module, the multi-dimensional spatiotemporal topology mapping engine, and the path integrity verification and cache management module, respectively. It is used to respond to the data change event and execute differentiated calculation strategies according to the consistency verification result. When the path integrity verification and cache management module confirms that the consistency verification is passed, the event-driven and incremental calculation module directly calls the calculation of the composite transmission coefficient credit potential energy of the cache. When the consistency verification result is that the verification fails, it performs hop-by-hop traversal calculation and updates the credit score of the affected node.

[0145] Specifically, the event-driven and incremental calculation module in this embodiment includes a change impact assessment unit, a potential energy transfer calculation unit, and a safety acceleration control unit. The event-driven and incremental calculation module is used to transform discrete data changes into continuous credit status updates in the graph, and dynamically selects a calculation strategy based on the path integrity verification results.

[0146] The change impact assessment unit is connected to the multi-source data acquisition and standardization module to receive data change events. First, identify the changes in features contained in the data change event. Mapped to initial credit potential shock This mapping is not a simple linear transformation, but rather based on a pre-defined feature sensitivity matrix. The weighted calculation is performed. The impact on the credit rating system differs for different types of characteristics (e.g., changes in registered capital and new administrative penalties). The formula for calculating the initial credit potential impact is defined as:

[0147] ;

[0148] In the formula, This is the standardized feature increment vector. This is a diagonal matrix, where the diagonal elements correspond to the risk sensitivity coefficients of each feature dimension. The calculated... This represents the original energy generated at the source node during this data change event, which will soon be injected into the multidimensional spatiotemporal topology map for diffusion.

[0149] The potential energy transmission calculation unit is used to simulate the physical transmission process of credit risk in the graph. A credit potential energy transmission model is introduced to quantify the diffusion effect of risk. In this model, each node in the graph is considered a potential energy storage container, and the directed edges between nodes are considered energy conduits. For any affected node in the graph... Its accumulated credit potential It is defined as the sum of the input energy from all incoming neighbor nodes. Incoming neighbor nodes refer to the upstream nodes that are directly pointed to by directed edges in the multidimensional spatiotemporal topology graph, including entity nodes and field nodes. Their output energy is the credit potential energy of that node.

[0150] To avoid invalid recalculation across the entire graph caused by minor numerical fluctuations, the potential energy transfer calculation unit sets a dynamic sensitivity threshold for each node. A node's state will only be activated and continue to propagate downstream when the accumulated potential energy exceeds this threshold. The logic for determining potential energy accumulation and activation is expressed as follows:

[0151] ;

[0152] In the formula, Represents a node Potential energy, Represents nodes Adjacent other nodes, Represents a node The set of neighboring nodes, Indicates adjacent nodes The output value, Indicates from node To the node The weighting coefficients.

[0153] like The node is then activated, generating new output potential and triggering a credit score update; if If the potential energy fluctuation at that node is then considered background noise and blocked, then... Represents a node The set of incoming neighbors, Energy transmitted from upstream nodes The edge weights are maintained by the multidimensional spatiotemporal topology mapping engine.

[0154] The security acceleration control unit is connected to the path integrity verification and cache management module, realizing a computation acceleration mechanism based on cryptographic verification (i.e., secure incremental computation). The credit potential energy calculation is prepared from the source node in the potential energy conduction computation unit. To the target node When the long link propagation is affected, the security acceleration control unit does not immediately perform hop-by-hop traversal, but instead sends a verification request for the path to the path integrity verification and cache management module and receives the returned consistency verification result.

[0155] Based on the consistency verification results, the security acceleration control unit dynamically switches between the following two computing modes:

[0156] In Mode 1, when the received consistency verification result is either verification passed or consistent, the security acceleration control unit enters the fast computation mode (corresponding to the fast computation channel). In this fast computation mode, the security acceleration control unit directly retrieves the cached composite conduction coefficient. The potential energy increment of the target node is calculated directly through single-step multiplication.

[0157] ;

[0158] In the formula, This represents the potential energy increment at the target node. Represents the initial credit potential. The composite conduction coefficient represents the target conduction path.

[0159] The operation will reduce the original complexity to O(n). ( The chained computation (for path length) is reduced to The constant-level operations significantly reduce computational latency. Because... Having already undergone consistency anchoring through the fingerprint chain, this leapfrog calculation improves speed while ensuring the accuracy and security of the values.

[0160] In Mode 2, when the received consistency verification result indicates verification failure or tampering, the security acceleration control unit initiates a rollback calculation mode (corresponding to a hop-by-hop traversal calculation channel). In this rollback calculation mode, the security acceleration control unit performs a breadth-first search (BFS) or depth-first search (DFS) traversal, calculating and accumulating potential energy hop-by-hop along the edge structure of the graph, while simultaneously triggering the reading of the latest state of nodes along the path. Although this mode has a higher computational cost, it ensures calculation accuracy in the event of cache failure or data anomalies and provides the latest baseline data for the next cache build.

[0161] Finally, the event-driven and incremental computation module updates the credit scores of affected nodes based on the calculated potential energy changes, and persists the updated results to the cold and hot tiered encrypted storage scheduler module. Through this mechanism, the system ensures high-concurrency computing performance while using an underlying encrypted verification mechanism to prevent credit evaluation biases caused by cache pollution or malicious tampering.

[0162] like Figure 8 As shown, the application service interface module is connected to the event-driven and incremental calculation module, and is used to receive external query requests, retrieve the updated credit score, and generate an encrypted credit report.

[0163] Specifically, the application service interface module in this embodiment includes a request gateway unit, a level assessment and report generation unit, and a security delivery unit.

[0164] The request gateway unit is configured as the system's API access point to receive query requests from financial institutions, regulatory authorities, or enterprise users. To prevent unauthorized access and replay attacks, the request gateway unit employs an authentication mechanism combining bidirectional transport layer security (mTLS) and API key authentication. When an external terminal initiates a query request, the request gateway unit verifies the digital signature and timestamp in the request header. After confirming the legitimacy of the request, it parses the unique identifier of the query target and distributes instructions to downstream modules.

[0165] The rating and report generation unit connects to the event-driven and incremental calculation module to receive the calculated credit score values ​​and perform a mapping from the continuous numerical space to the discrete credit rating space. Although the underlying credit potential calculation provides a high-precision floating-point score, in specific credit approval or business decision-making scenarios, decision-makers rely more on intuitive credit ratings.

[0166] The credit rating and report generation unit incorporates a dynamic quantile mapping algorithm, which determines the boundary thresholds for credit rating based on the score distribution of all credit subjects. For any target subject's credit score... Its corresponding credit rating Determined by the following piecewise mapping function:

[0167] ;

[0168] In the formula, Indicates credit rating. This indicates the current credit score. This represents the threshold for credit scoring.

[0169] Using dynamic percentiles instead of fixed values ​​as thresholds automatically adapts to overall score drift caused by economic cycle fluctuations, ensuring the statistical stability of credit rating distribution. After determining the credit rating, the rating and report generation unit calls a preset report template to assemble the credit rating, base score, main risk attributions (i.e., key negative factors that lead to a decrease in the score), and related graph subgraph structures into a structured credit report document.

[0170] The Secure Delivery Unit is responsible for performing secondary encryption before report output, building the last line of defense for data before it leaves the system boundary. Although the system's internal storage already employs encryption mechanisms, to prevent reports from being intercepted or tampered with by men in the middle during network transmission, the Secure Delivery Unit enforces an end-to-end encryption strategy based on digital envelopes.

[0171] Specifically, after the credit report is generated, the secure delivery unit randomly generates a one-time symmetric session key. The session key is then used to perform high-strength symmetric encryption on the credit report content, generating the ciphertext report. Subsequently, the secure delivery unit obtains the public key reserved by the query initiator during registration. Use the public key to set the session key Asymmetric encryption is performed to generate the key ciphertext. Finally, the secure delivery unit packages the report ciphertext and the key ciphertext together and returns it as the final response message to the external terminal. Upon receiving the message, the external terminal can only decrypt it using its private key to obtain the session key, and then decrypt and read the credit report.

[0172] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A multi-dimensional credit data processing system for cloud architecture, characterized in that, include: The multi-source data acquisition and standardization module is used to map raw data into standardized feature vectors and monitor changes in standardized feature vectors to generate data change events. A multidimensional spatiotemporal topology mapping engine is used to receive the standardized feature vectors and construct a multidimensional spatiotemporal topology map. The query intent analysis and topology warm-up module is used to generate preloading instructions for the corresponding subgraph range based on the target query pattern. The hot and cold tiered encrypted storage scheduler module is used to schedule node data between the hot zone of the memory database and the cold zone of the object storage according to the preloading instruction, and to encrypt the node data that sinks to the cold zone of the object storage to generate encrypted data objects. The path integrity verification and cache management module constructs a cache fingerprint using encrypted data objects corresponding to each node on the target transmission path based on the multidimensional spatiotemporal topology map. Before the credit potential calculation is initiated, it reads the encrypted data objects of each node to generate a real-time fingerprint and compares the real-time fingerprint with the cache fingerprint to generate a verification result. The event-driven and incremental calculation module is used to respond to the data change event, calculate the credit potential based on the verification result according to the credit potential transmission model, obtain the potential change, and update the credit score.

2. The system according to claim 1, characterized in that, The multi-source data acquisition and standardization module is specifically used for: The original data is converted into a standardized feature vector using a preset standardization function; The data change event includes the credit subject's unique identifier, the amount of feature change, the timestamp, and the operation type.

3. The system according to claim 1, characterized in that, The multidimensional spatiotemporal topology mapping engine is also used to calculate the weights of the associated edges and belonging edges between nodes; The multidimensional spatiotemporal topology mapping engine constructs field nodes in the following ways: Parse regional keywords from the address field of the credit subject and materialize the regional keywords into independent nodes in the graph; The multidimensional spatiotemporal topology mapping engine establishes the belonging edges of entity nodes to the field nodes.

4. The system according to claim 1, characterized in that, The query intent analysis and topology warm-up module calculates the frequency of the target query pattern using a sliding time window statistical model.

5. The system according to claim 4, characterized in that, The calculation logic for the frequency of the target query pattern is as follows: Within the sliding window, the requests matching the target query pattern are weighted and summed, then divided by the window duration. When the frequency of the calculated target query pattern exceeds a preset popularity threshold, the preloading instruction is generated.

6. The system according to claim 1, characterized in that, The logic configuration for data encryption and decryption performed by the cold and hot tiered encrypted storage scheduler module is as follows: When the node data needs to be persisted to the object storage cold zone, the serialized node data is encrypted using a preset platform public key to generate the encrypted data object; When the node data needs to be loaded into the hot zone of the memory database, the encrypted data object is decrypted and restored using the platform private key that is only stored in the memory isolation zone.

7. The system according to claim 1, characterized in that, The path integrity verification and cache management module constructs a topological hash fingerprint chain as a cache fingerprint and stores it in the hot zone of the in-memory database. The construction method includes: Obtain the encrypted data object corresponding to each node on the target transmission path in the object storage cold zone; Perform hash digest calculation on the encrypted data object to generate a node-level fingerprint; According to the topological order of the target transmission path, the node-level fingerprints of all nodes on the target transmission path are concatenated and spliced ​​together. A hash operation is performed again on the concatenated data to generate a unique fingerprint chain signature for the target transmission path.

8. The system according to claim 3, characterized in that, The path integrity verification and cache management module pre-calculates the composite transmission coefficient of the target transmission path, which is equal to the product of the weights of all edges on the target transmission path.

9. The system according to claim 3, characterized in that, The credit potential transmission model defined in the event-driven and incremental calculation module includes: multiplying the feature change amount in the data change event with the feature sensitivity matrix to obtain the initial credit potential; The initial credit potential energy is injected into the multidimensional spatiotemporal topology graph as the output energy of the source node. For any affected node in the multidimensional spatiotemporal topology graph, its accumulated credit potential energy is calculated. When the accumulated credit potential exceeds the dynamic sensitivity threshold, the node's credit score is updated and the credit potential is transmitted to downstream nodes.

10. The system according to claim 1, characterized in that, Also includes: The application service interface module is used to map the updated credit score to a credit rating.

Images