A cloud computing-based risk control credit monitoring method

Abstract

The application discloses a kind of based on cloud computing's risk control credit monitoring method, belong to cloud computing technical field, a kind of based on cloud computing's risk control credit monitoring method includes the following steps: S1. real-time acquisition user transaction data and behavior trajectory data;S2. data is washed and standardized processing;S3. based on machine learning constructs multidimensional risk assessment model;S4. according to risk characteristics dynamic generation credit score;S5. to abnormal transaction real-time trigger early warning mechanism;S6 generates visual risk control report and updates monitoring strategy.The application integrates multi-source heterogeneous data through federated learning, combines homomorphism encryption and differential privacy to realize hierarchical privacy protection, uses hybrid cloud resource scheduling and dynamic model updating technology to improve the real-time performance of risk assessment, and builds compliance audit closed loop based on blockchain and explainability analysis.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of cloud computing, and particularly relates to a risk control credit monitoring method based on cloud computing. BACKGROUND

[0002] The traditional risk control credit monitoring method faces the following key bottlenecks in the cloud computing and multi-institution cooperation scene, which specifically shows that: data island and inefficient heterogeneous integration, the existing method usually only relies on the structured data of a single institution and lacks deep analysis ability of unstructured text; privacy protection means is fragmented, the existing scheme adopts isolated technology, such as homomorphic encryption only protects static data, and the federal learning framework does not cover the gradient leakage risk; mixed cloud resource scheduling is rigid, the traditional mixed cloud architecture does not schedule tasks according to the data sensitivity, resulting in that the high security requirement calculation is forced to be deployed in the public cloud, or the non-sensitive task occupies the private cloud resource; compliance audit and decision black box, the existing method relies on offline log audit, lacks real-time evidence and explainability support, such as the risk score of the black box model (such as deep learning) cannot be traced back to specific data features, and it is difficult to review through the GDPR "explanation right" clause.

[0003] In view of the above problems, based on the core innovation of cross-modal data deep fusion mechanism, three-level privacy protection system, intelligent mixed cloud scheduling strategy and dynamic compliance closed loop, under the premise of guaranteeing data privacy and compliance, the accurate construction and real-time iteration of the cross-institution risk control model are realized, and a safe, efficient and transparent credit monitoring solution is provided for the financial technology field. SUMMARY

[0004] The application discloses a risk control credit monitoring method based on cloud computing, which integrates multi-source heterogeneous data through federal learning, realizes hierarchical privacy protection by combining homomorphic encryption and differential privacy, improves the real-time performance of risk assessment by using mixed cloud resource scheduling and dynamic model updating technology, and constructs a compliance audit closed loop based on blockchain and explainability analysis.

[0005] The technical scheme of the application is implemented as follows:

[0006] The application discloses a risk control credit monitoring method based on cloud computing, which includes the following steps:

[0007] S1. Real-time collection of user transaction records, credit history and enterprise financial data through a structured data unit to generate structured data labels; integration of an NLP natural language processing engine in unstructured text units to analyze free text in social media, enterprise announcements and judicial documents, extract entity keywords and generate text feature labels; in a cross-institutional scenario, the federal learning client divides the multi-source data according to the institution domain and generates an encrypted feature index to provide metadata for distributed training;

[0008] S2. Based on the encryption feature index generated in S1, the sensitive data is processed using a homomorphic encryption algorithm to support ciphertext transmission, storage, and partial calculation; the federated learning framework aligns the feature dimensions of the multi-source encrypted data to generate cross-domain joint training intermediate parameters; the dynamic access control unit generates a dynamic token based on the unique hash value generated by the user role and device hardware information, and authorizes entities to access the desensitized data subset within a specified time window;

[0009] S3. Based on the intermediate parameters in S2 and the desensitized data, the private cloud node stores the core sensitive data and performs redundant backup, and calls the public cloud elastic resource to perform non-sensitive feature extraction and model training; the federated learning coordinator divides the training task into local subtasks and global aggregation tasks; the task distribution controller allocates the calculation task based on the structured data label, text feature label generated in S1, and the public cloud resource state, and routes the sensitive calculation to the private cloud node through the private cloud gateway strategy;

[0010] S4. Based on the local subtask training results in S3, the rule engine calls the risk control rule library to output risk events and confidence, and the machine learning unit aggregates the local base models of multiple participants to build a global scoring model through weighted averaging or model integration method; the integrated SHAP analysis unit generates a feature contribution report, and the local differential privacy mechanism adds noise to the model gradient to obscure the sensitive features; the outputs of the rule engine and the model unit are fused by weighting to generate a risk score, and the weight is dynamically adjusted according to the real-time calculation of the model accuracy and recall rate;

[0011] S5. Based on the risk score in S4, the stream data engine receives the user transaction records in S1 and the risk score in S4, processes the user behavior sequence to generate a dynamic portrait; the adaptive update unit monitors the model decay indicators to trigger incremental learning to update the model parameters; during the asynchronous update process in the federated framework, the parameter server periodically aggregates the global model parameters, and the federated learning coordinator manages task splitting and resource allocation;

[0012] S6. Aggregate the text features in S1, the risk indicators in S4, the dynamic portrait in S5, and the federated learning operation records to build a user-enterprise-transaction multi-dimensional relationship network and render a risk heat map; the compliance audit unit generates a time-stamped operation chain through blockchain storage, analyzes the mapping relationship between the federated learning records and the privacy agreement clauses, and outputs an audit report integrating the GDPR general data protection regulation, the PIPL personal information protection law evaluation items, and the traceable matrix based on the blockchain operation chain; the independent verification module verifies the compliance of the identity authentication information, gradient transmission integrity, and differential privacy noise addition.

[0013] Preferably, S1 specifically comprises the following steps:

[0014] S1-1. The structured data unit collects the transaction data of the financial institutions in real time through the distributed message queue, uses the pattern matching technology based on regular expressions to clean up the missing values and abnormal values in the data, uses the normalization method to map the numerical features to the standardized interval, generates the structured data label containing the account identifier, transaction timestamp and amount dimension, and the structured data label is stored in the column storage format;

[0015] S1-2. The unstructured text analysis module loads the pre-trained language model to perform semantic analysis on the judicial documents and enterprise announcements, identifies the legal entities and financial indicators in the text through the bidirectional neural network architecture, stores the extraction results in the graph structure data format, and establishes the entity association mapping with the structured data;

[0016] S1-3. The cross-institutional federated data directory constructs a distributed index based on the encrypted hash table, aligns the spatial features of the multi-source heterogeneous data, extracts the data distribution features through the deep neural network to generate the encrypted metadata, and the metadata update uses the decentralized synchronization protocol to ensure consistency.

[0017] Preferably, the S2 specifically comprises the following steps:

[0018] S2-1. The homomorphic encryption module performs polynomial transformation operation on the sensitive fields, completes the feature cross operation in the ciphertext space, generates the verifiable encrypted feature matrix, and the matrix dimension is compressed and optimized through the linear algebra method;

[0019] S2-2. The feature hashing process maps the multi-source data to a unified vector space using a non-collision hash function, calculates the statistical distribution features in the hash bucket to generate the joint training intermediate parameters, and the parameter serialization uses the compact binary format to improve the transmission efficiency;

[0020] S2-3. The dynamic token generator fuses the device hardware fingerprint and digital certificate information, generates the access credential with time limit through the hash message authentication code algorithm, binds the device geofence and implements the time window verification mechanism.

[0021] Preferably, the S3 specifically comprises the following steps:

[0022] S3-1. The private cloud storage cluster is deployed using the distributed file system architecture, realizes the redundant storage of data blocks based on the erasure code algorithm, integrates the hardware-level encryption module in the storage process to protect the data security, and the key life cycle management is realized through the special security device;

[0023] S3-2. The hybrid cloud resource scheduler monitors the computing load indicators of the graphics processing unit, dynamically allocates the feature processing tasks to the acceleration chips of different architectures according to the preset strategy, and implements the load balancing and failover mechanism in the task distribution process;

[0024] S3-3. The trusted task routing controller enables a hardware isolated protection environment for high-sensitivity tasks according to the data security level label, establishes an encrypted communication channel conforming to the transmission layer security standard, and adopts an intelligent path selection algorithm for network routing strategy.

[0025] Preferably, the S4 specifically includes the following steps:

[0026] S4-1. The rule engine loads an anti-fraud rule library containing account abnormal behavior patterns, adopts the Rete algorithm of the Drools rule engine for multi-condition reasoning matching, the rule conditions include risk event feature combinations such as sudden increase in transaction frequency, cross-regional device login, and large amount of transfer payee black list, and the matching successful risk event triggers an alarm work order generation process;

[0027] S4-2. The federated model aggregator receives local model parameter updates from each participant through a secure multi-party computation protocol, adopts a weighted average algorithm to globally aggregate the hidden layer weight matrix of the credit scoring model, and uses bilinear mapping to verify parameter integrity and authenticity during the aggregation process;

[0028] S4-3. The differential privacy module injects random noise that satisfies the probability density distribution into the gradient transmission before the model update, and the noise parameter is dynamically adjusted according to the data sensitivity level, ensuring that the intermediate parameters in the training process cannot be inversely deduced from the original training samples.

[0029] Preferably, the S5 specifically includes the following steps:

[0030] S5-1. The user portrait generator processes time series transaction behavior data through a long short-term memory neural network, the network structure includes a bidirectional recurrent layer and an attention mechanism, and outputs a dynamic feature vector representing consumption periodicity and fund flow pattern, the feature vector dimension is automatically aligned with the input layer of the credit scoring model;

[0031] S5-2. The model decay monitor continuously calculates the relative change rate of the area under the ROC curve of the validation set, and when the model prediction performance is monitored to decrease by more than a preset threshold, triggers an incremental training task containing samples of the latest time period, and the incremental training adopts an elastic weight consolidation algorithm to prevent catastrophic forgetting;

[0032] S5-3. The asynchronous update coordinator adopts a historical gradient momentum buffer mechanism to optimize the parameter synchronization process, the coordinator maintains an independent version control queue for each participant, and determines the parameter fusion ratio by comparing the cosine similarity of the global model and the local model.

[0033] Preferably, the S6 specifically includes the following steps:

[0034] S6-1. The graph computing engine constructs a heterogeneous relationship graph containing enterprise shareholder chains, guarantee relationships and actual controller paths, adopts a random walk-based graph embedding algorithm to calculate node centrality scores, and a risk propagation model identifies hidden associated party credit risks by simulating fund flow directions;

[0035] S6-2. The blockchain storage module organizes data operation records in chronological order into a hash chain structure, each block containing operation type, executor digital identity and operation object fingerprint information, and storage data achieves distributed ledger consistency through a Byzantine fault-tolerant consensus algorithm;

[0036] S6-3. The compliance verifier verifies cross-domain data transmission paths using a non-interactive zero-knowledge proof protocol, the prover generates an evidence chain containing data hash values and routing nodes, and the verifier confirms that no man-in-the-middle tampering has occurred in the transmission process through elliptic curve bilinear pair operations.

[0037] Compared with the prior art, the advantages of the present application are:

[0038] (1) Multi-source heterogeneous data deep fusion: structured data units and unstructured text units work together, extract entity keywords and generate feature labels through an NLP engine, and realize cross-institutional data domain alignment through a federated learning client, breaking down data silos and improving risk assessment dimensions;

[0039] (2) Hierarchical privacy protection design: combine homomorphic encryption (data level), local differential privacy (gradient level) and federated learning (collaboration level) to form a multi-level protection system; dynamic access control based on device fingerprint and time token ensures minimum access, balancing security and efficiency;

[0040] (3) Hybrid cloud elastic resource scheduling: through task distribution controller and private cloud gateway strategy, sensitive computing is forced to route to private cloud, non-sensitive tasks are flexibly distributed to public cloud, optimizing resource utilization; redundancy backup and erasure code technology guarantee data high availability. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 The flowchart of the overall method of the present application. DETAILED DESCRIPTION

[0042] EMBODIMENT

[0043] As Figure 1 shown, the present application discloses a cloud computing-based risk control credit monitoring method, comprising the following steps:

[0044] S1. Real-time collection of user transaction records, credit history and enterprise financial data by structured data unit to generate structured data labels; integration of NLP natural language processing engine by unstructured text unit to parse free text in social media, corporate announcements and judicial documents, extract entity keywords and generate text feature labels; in cross-institutional scenarios, federated learning clients divide multi-source data by institution domain and generate encrypted feature index to provide metadata for distributed training;

[0045] The S1 specifically comprises the following steps:

[0046] S1-1. The structured data unit collects transaction stream data of financial institutions in real time through a distributed message queue, uses regular expression-based pattern matching technology to clean up missing values and abnormal values in the data, uses normalization methods to map numerical features to a standardized interval, and generates structured data labels containing account identifiers, transaction timestamps and amount dimensions. The structured data labels are stored in columnar storage format for persistent storage;

[0047] S1-2. The unstructured text analysis module loads a pre-trained language model to perform semantic analysis on judicial documents and corporate announcements, identifies legal entities and financial indicators in the text through a bidirectional neural network architecture, and stores the extracted results in a graph structure data format, and establishes an entity association mapping with the structured data;

[0048] S1-3. Cross-institutional federated data directory based on encrypted hash table to build distributed index, spatially align multi-source heterogeneous data features, extract data distribution features through deep neural network to generate encrypted metadata, and use a decentralized synchronization protocol to ensure consistency of metadata updates, which specifically includes the following: clear fault tolerance mechanism and conflict resolution rules, add generative adversarial network and contrastive learning technology:

[0049] S1-3-1. Distributed index construction: the encrypted hash table uses an anti-collision hash function based on the SM3 national encryption algorithm to generate index key values, and each key value corresponds to the logical address of the data shard in the institution domain; the index key values are matched through a homomorphic encryption algorithm (Paillier encryption) in a fuzzy manner in a cross-institutional ciphertext state, supporting privacy-protected joint queries;

[0050] S1-3-2. Spatial alignment operation: use the generative adversarial network (GAN) to train the multi-source data feature space projector to map heterogeneous data from different institutions to a unified semantic space; the alignment error is optimized through a contrastive learning loss function (InfoNCE) to ensure that the cosine similarity of feature vectors of the same entity in different institution domains is ≥0.85;

[0051] S1-3-3. The deep neural network adopts a residual convolution structure, the input is the data distribution histogram and statistical features in the domain of the organization, and the output layer generates a 128-dimensional encrypted metadata vector; the metadata vector is stored in multiple federal nodes through a threshold secret sharing algorithm, and any node cannot restore the original data distribution alone;

[0052] S1-3-4. Decentralized synchronization protocol: using an improved Raft consensus algorithm (supporting Byzantine fault tolerance), initiating a pre-commit vote when updating metadata, and the node needs to verify the consistency of the data hash value and digital signature; during synchronization, version conflicts are resolved through a timestamp-based vector clock mechanism, and the latest operation record is preferentially retained;

[0053] S2. Based on the encrypted feature index generated by S1, homomorphic encryption algorithm is used to process sensitive data, supporting ciphertext transmission, storage and partial calculation; the federated learning framework aligns the feature dimensions of multi-source encrypted data, generating cross-domain joint training intermediate parameters; the dynamic access control unit generates a dynamic token based on the unique hash value generated by the user role and device hardware information, and the authorized entity accesses the desensitized data subset within a specified time window;

[0054] The S2 specifically comprises the following steps:

[0055] S2-1. The homomorphic encryption module performs a polynomial transformation operation on the sensitive field, completes the feature cross operation in the ciphertext space, generates a verifiable encrypted feature matrix, and the matrix dimension is compressed and optimized through linear algebra method, which specifically includes the following contents:

[0056] Explicit homomorphic encryption algorithm and parameter configuration, and add zero-knowledge proof verification mechanism:

[0057] S2-1-1. Polynomial transformation operation: RLWE (Ring Learning with Errors) homomorphic encryption scheme is used for sensitive fields (such as ID number and transaction amount) to map plaintext to ciphertext coefficients on the polynomial ring; the polynomial order is set to 2048, the modulus q is a safe prime number (q = 12289), and the noise distribution uses discrete Gaussian distribution (standard deviation σ = 3.19);

[0058] S2-1-2. Ciphertext feature cross operation: perform feature cross multiplication operation in ciphertext state, use polynomial multiplication to realize cross item generation, and the operation result is decomposed into low-dimensional polynomial group through Chinese remainder theorem (CRT); cross item verification uses zero-knowledge proof protocol (zk-SNARK) to generate verifiable ciphertext operation correctness proof;

[0059] S2-1-3. Matrix compression optimization: The encrypted feature matrix is approximated by low rank using singular value decomposition (SVD), and the first k singular values are retained (k = min(m, n) x 0.2), and the threshold of truncation error is set to 1e-4; the compressed matrix is stored in sparse matrix format, and the non-zero elements are further compressed in storage space by run-length encoding (RLE);

[0060] S2-2. The feature hashing process uses a non-collision hash function to map multi-source data to a unified vector space, calculates the statistical distribution features in the hash bucket to generate joint training intermediate parameters, and serializes the parameters in a compact binary format to improve transmission efficiency;

[0061] S2-3. The dynamic token generator combines device hardware fingerprints and digital certificate information to generate time-sensitive access credentials through a hash message authentication code algorithm, and the access credentials are bound to the device geofence and implement a time window verification mechanism;

[0062] S3. Based on the intermediate parameters and desensitized data of S2, the private cloud node stores the core sensitive data and performs redundant backup, and calls the public cloud elastic resource to perform non-sensitive feature extraction and model training; the federal learning coordinator divides the training task into local subtasks and global aggregation tasks; the task distribution controller assigns the calculation tasks based on the structured data labels, text feature labels and public cloud resource states generated by S1, and routes the sensitive calculation to the private cloud node through the private cloud gateway strategy;

[0063] S3 specifically includes the following steps:

[0064] S3-1. The private cloud storage cluster is deployed based on a distributed file system architecture, and the data block redundancy storage is realized based on an erasure code algorithm, the storage process integrates a hardware-level encryption module to protect data security, and the key life cycle management is realized through a special security device;

[0065] S3-2. The hybrid cloud resource scheduler monitors the computing load indicators of the graphics processing unit, dynamically allocates feature processing tasks to different architecture acceleration chips according to the preset strategy, and implements load balancing and failover mechanism in the task distribution process;

[0066] S3-3. The trusted task routing controller enables hardware isolation protection environment for high-sensitive tasks according to data security level labels, establishes an encrypted communication channel conforming to the transmission layer security standard, and adopts an intelligent path selection algorithm for network routing strategy;

[0067] S4. Based on the local sub-task training results of S3, the rule engine calls the risk control rule library to output risk events and confidence, the machine learning unit aggregates the local base models of multiple participants, and constructs a global scoring model through weighted averaging or model integration method; the SHAP analysis unit generates a feature contribution report, and the local differential privacy mechanism adds noise to the model gradient to blur the sensitive features; the outputs of the rule engine and the model unit are fused through weighted fusion to generate a risk score, and the weight is dynamically adjusted according to the real-time calculated model accuracy and recall rate;

[0068] The S4 specifically comprises the following steps:

[0069] S4-1. The rule engine loads the anti-fraud rule library containing account abnormal behavior patterns, adopts the Rete algorithm of the Drools rule engine for multi-condition reasoning matching, the rule conditions include risk event feature combinations such as sudden increase in transaction frequency, cross-regional device login, and large amount of transfer payee blacklist, and the matching successful risk event triggers the alarm work order generation process;

[0070] S4-2. The federated model aggregator receives the local model parameter updates of each participant through the secure multi-party computation protocol, adopts the weighted average algorithm to globally aggregate the hidden layer weight matrix of the credit scoring model, and adopts the bilinear mapping to verify the parameter integrity and source authenticity during the aggregation process, which specifically includes the following contents:

[0071] Mathematical basis for explaining dynamic weight adjustment:

[0072] S4-2-1. Secure multi-party computation protocol: adopt Shamir secret sharing scheme, each participant divides the local model parameters into n fragments (n≥3t+1, t is the maximum fault tolerance node number), and distributes them to other participants; the aggregator reconstructs the global parameters through Lagrange interpolation method, and implements threshold signature mechanism to verify the legality of the fragment source during the process;

[0073] S4-2-2. Weighted average algorithm: the weight is calculated according to the F1-score of the local model on the validation set, and the dynamic adjustment formula is:

[0074]

[0075] Wherein, the hidden layer weight matrix aggregation adopts block parallel calculation, each matrix block independently executes weighted average operation, and the block size is set to 256x256;

[0076] F1 i : F1-score (harmonic mean of precision and recall) of the i-th participant's local model on the validation set; N: total number of participants in federated learning- square operation (F1 2) for amplifying the weight contribution of high-precision models; the denominator is the sum of the squares of all participating F1-scores, ensuring weight normalization;

[0077] S4-2-3. Bilinear mapping verification: define a bilinear pair function e: G1 x G2→ G T , where G1, G2 are elliptic curve groups, and the participants need to provide group element commitments of model parameters;

[0078] Verify the equation:

[0079]

[0080] where H(W i ) is the BLAKE2 hash value of the weight matrix, and verification failure triggers the parameter rollback mechanism; g1, g2: the generators of groups G1 and G2, respectively;

[0081] S4-3. Differential privacy module injects random noise that satisfies the probability density distribution before gradient transmission, and the noise parameter is dynamically adjusted according to the data sensitivity level, ensuring that the intermediate parameters in the training process cannot be inversely deduced from the original training samples;

[0082] S5. Risk score based on S4, the stream data engine receives the user transaction records of S1 and the risk score of S4, processes the user behavior sequence to generate a dynamic portrait; the adaptive update unit monitors the model decay indicators and triggers incremental learning to update the model parameters; in the asynchronous update process of the federal framework, the parameter server periodically aggregates the global model parameters, and the federal learning coordinator manages task splitting and resource allocation;

[0083] The S5 specifically comprises the following steps:

[0084] S5-1. The user portrait generator processes time-series transaction behavior data through a long short-term memory neural network, the network structure includes a bidirectional recurrent layer and an attention mechanism, and outputs a dynamic feature vector representing consumption periodicity and fund flow patterns. The feature vector dimension is automatically aligned with the input layer of the credit scoring model;

[0085] S5-2. The model decay monitor continuously calculates the relative change rate of the area under the ROC curve of the validation set, and when the model prediction performance is found to have decreased by more than a preset threshold, triggers an incremental training task containing samples from the latest time period. Incremental training uses an elastic weight consolidation algorithm to prevent catastrophic forgetting;

[0086] S5-3. The asynchronous update coordinator uses a historical gradient momentum buffer mechanism to optimize the parameter synchronization process, and maintains an independent version control queue for each participant. The coordinator determines the parameter fusion ratio by comparing the cosine similarity of the global model and the local model;

[0087] S6. Aggregating the text features of S1, the risk indicators of S4, the dynamic portraits of S5 and the federated learning operation records, constructing a user-enterprise-transaction multi-dimensional relationship network and rendering a risk heat map; the compliance audit unit generates a time-stamped operation chain through blockchain storage, analyzes the mapping relationship between federated learning records and privacy protocol clauses, and outputs an audit report integrating GDPR general data protection regulation, PIPL personal information protection law evaluation items and a traceable matrix based on the blockchain operation chain; the independent verification module verifies the compliance of identity authentication information, gradient transmission integrity and differential privacy noise addition;

[0088] The S6 specifically includes the following steps:

[0089] S6-1. The graph computing engine constructs a heterogeneous relationship graph containing enterprise shareholder chains, guarantee relationships and actual controller paths, calculates node centrality scores using a graph embedding algorithm based on random walk, and identifies hidden associated party credit risks through a risk propagation model by simulating fund flow, specifically including the following contents: defining graph walk parameters and centrality calculation formula, supplementing post-enhancement technology to improve repeatability:

[0090] S6-1-1. Heterogeneous relationship graph construction: define three types of edge attributes: shareholder relationship (holding ratio ≥ 5%), guarantee relationship (joint and several liability type), and controller path (penetrating holding level ≥ 3 layers); node attribute fusion of S4 risk score and S5 dynamic portrait features, attribute missing values are interpolated and completed by graph convolution network (GCN);

[0091] S6-1-2. Graph embedding algorithm optimization: use Node2Vec algorithm to generate node embedding vectors, bias parameter setting p=0.5 (breadth-first), q=2.0 (depth-first), walk length 80 steps;

[0092] Improved centrality score calculation formula:

[0093] C(v) = a PageRank(v) + β Betweenness(v)

[0094] Where a = 0.6, β = 0.4, and the parameter combination is optimized by grid search;

[0095] S6-1-3. Risk propagation model: based on Monte Carlo simulation of fund flow path, the initial risk source is the high-risk enterprise node marked by S4, the propagation probability formula is:

[0096] P u→v =1-exp(-λ·w uv ·R u )

[0097] Where, λ is the attenuation factor (λ = 0.85), w uvis an edge weight, R u is a risk score of node u;

[0098] The hidden association identification adopts a subgraph isomorphism detection algorithm (VF2 algorithm) to match a minimum support tree structure of a known risk pattern;

[0099] S6-2. The blockchain storage module organizes the data operation records into a hash chain structure in chronological order, each block contains operation type, executor digital identity and operation object fingerprint information, and the storage data achieves distributed ledger consistency through a Byzantine fault tolerance consensus algorithm;

[0100] S6-3. The compliance verifier verifies the cross-domain data transmission path by using a non-interactive zero-knowledge proof protocol, the prover generates an evidence chain containing data hash value and routing node, and the verifier confirms that the transmission process has not been tampered with by an intermediate person through an elliptic curve bilinear pair operation.

[0101] The above shows and describes the basic principles, main features and advantages of the present application. It should be understood by those skilled in the art that the present application is not limited by the above examples, the above examples and descriptions in the specification are only preferred examples of the present application, and are not intended to limit the present application, various changes and improvements can be made to the present application without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A cloud computing-based risk control credit monitoring method, characterized in that, Comprise the following steps: S1. Collect user transaction records, credit history and enterprise financial data in real time through structured data units to generate structured data labels; integrate NLP natural language processing engine to parse free text in social media, corporate announcements and judicial documents through unstructured text units to extract entity keywords and generate text feature labels; in cross-agency scenarios, the federal learning client divides multi-source data by agency domain and generates encrypted feature index to provide metadata for distributed training; S2. Based on the encrypted feature index generated in S1, use homomorphic encryption algorithm to process sensitive data to support ciphertext transmission, storage and partial calculation; the federal learning framework aligns the feature dimensions of multi-source encrypted data to generate intermediate parameters for cross-domain joint training; the dynamic access control unit generates a dynamic token based on the unique hash value generated by the user role and device hardware information to authorize entities to access the desensitized data subset within a specified time window; S3. Based on the intermediate parameters and desensitized data in S2, private cloud nodes store core sensitive data and perform redundant backup, and call public cloud elastic resources to perform non-sensitive feature extraction and model training; The federal learning coordinator splits the training task into local subtasks and global aggregation tasks; the task distribution controller assigns computing tasks based on the structured data labels, text feature labels and public cloud resource status generated in S1, and routes sensitive computing to private cloud nodes through private cloud gateway policies; S4. Based on the local subtask training results in S3, the rule engine calls the risk control rule library to output risk events and confidence, and the machine learning unit aggregates local base models from multiple participants to build a global scoring model through weighted averaging or model integration methods; the integrated SHAP analysis unit generates a feature contribution report, and the local differential privacy mechanism adds noise to the model gradient to obscure sensitive features; the outputs of the rule engine and the model unit are combined through weighted fusion to generate a risk score, and the weights are dynamically adjusted according to the real-time calculation of model accuracy and recall rate; S5. Based on the risk score in S4, the stream data engine receives user transaction records in S1 and risk scores in S4 to process user behavior sequences to generate dynamic portraits; The adaptive update unit monitors model decay indicators to trigger incremental learning to update model parameters; during asynchronous updating in the federal framework, the parameter server periodically aggregates global model parameters, and the federal learning coordinator manages task splitting and resource allocation; S6. Aggregate the text features in S1, risk indicators in S4, dynamic portraits in S5 and federal learning operation records to build a user-enterprise-transaction multi-dimensional relationship network and render a risk heat map; The compliance audit unit generates a time-stamped operation chain through blockchain storage, analyzes the mapping relationship between federal learning records and privacy protocol clauses, and outputs an audit report integrating GDPR general data protection regulations, PIPL personal information protection law evaluation items and traceable matrix based on the blockchain operation chain; the independent verification module verifies the compliance of identity authentication information, gradient transmission integrity and differential privacy noise addition.

2. The cloud computing-based risk control credit monitoring method according to claim 1, characterized in that, The S1 specifically comprises the following steps: S1-1. The structured data unit collects the transaction data of financial institutions in real time through a distributed message queue, uses a regular expression-based pattern matching technique to clean up missing values and abnormal values in the data, uses a normalization method to map numerical features to a standardized interval, generates structured data labels containing account identifiers, transaction timestamps, and amount dimensions, and stores the structured data labels in a columnar storage format for persistent storage; S1-2. The unstructured text analysis module loads a pre-trained language model to perform semantic analysis on judicial documents and enterprise announcements, identifies legal entities and financial indicators in the text through a bidirectional neural network architecture, stores the extraction results in a graph structure data format, and establishes an entity association mapping with the structured data; S1-3. The cross-institutional federated data directory constructs a distributed index based on a cryptographic hash table, aligns the features of multi-source heterogeneous data in space, extracts data distribution features through a deep neural network to generate encrypted metadata, and uses a decentralized synchronization protocol to ensure consistency in metadata updates. 3.The cloud computing-based risk control credit monitoring method of claim 1, wherein, The S2 specifically includes the following steps: S2-1. The homomorphic encryption module performs polynomial transformation operations on sensitive fields, completes feature cross operations in the ciphertext space, generates a verifiable encrypted feature matrix, and compresses and optimizes the matrix dimensions through linear algebra methods; S2-2. The feature hashing process uses a non-collision hash function to map multi-source data to a unified vector space, calculates the statistical distribution features in the hash bucket to generate joint training intermediate parameters, and uses compact binary format for parameter serialization to improve transmission efficiency; S2-3. The dynamic token generator fuses device hardware fingerprints and digital certificate information, generates time-sensitive access credentials through a hash message authentication code algorithm, binds the access credentials to a device geofence and implements a time window verification mechanism.

4. The cloud computing-based risk control credit monitoring method of claim 1, wherein, The S3 specifically includes the following steps: S3-1. The private cloud storage cluster is deployed using a distributed file system architecture, implements redundant storage of data blocks based on erasure coding algorithms, integrates a hardware-level encryption module during the storage process to protect data security, and manages the key lifecycle through a dedicated security device; S3-2. The hybrid cloud resource scheduler monitors the computational load indicators of the graphics processing unit, dynamically allocates feature processing tasks to different architecture acceleration chips based on pre-set strategies, and implements load balancing and failover mechanisms during task distribution; S3-3. The trusted task routing controller enables hardware isolation protection environment for high-sensitive tasks based on data security level labels, establishes an encrypted communication channel that complies with the Transport Layer Security standard, and uses an intelligent path selection algorithm for network routing strategy.

5. The cloud computing-based risk control credit monitoring method according to claim 1, characterized in that, The S4 specifically includes the following steps: S4-1. The rule engine loads an anti-fraud rule library containing account abnormal behavior patterns, uses the Rete algorithm of the Drools rule engine for multi-condition reasoning matching, the rule conditions include transaction frequency surge, cross-regional device login, and large amount of transfer recipient blacklist risk event feature combination, and the matching successful risk event triggers the alarm work order generation process; S4-2. The federal model aggregator receives local model parameter updates from each participant through a secure multi-party computation protocol, uses a weighted average algorithm to globally aggregate the hidden layer weight matrix of the credit scoring model, and uses bilinear mapping to verify parameter integrity and authenticity during the aggregation process; S4-3. The differential privacy module injects random noise that satisfies the probability density distribution before the gradient transmission, and the noise parameter is dynamically adjusted according to the data sensitivity level to ensure that the intermediate parameters in the training process cannot be inversely deduced from the original training samples.

6. The cloud computing-based risk control credit monitoring method according to claim 1, characterized in that, The S5 specifically comprises the following steps: S5-1. The user portrait generator processes time-series transaction behavior data through a long short-term memory neural network, the network structure includes a bidirectional recurrent layer and an attention mechanism, and outputs a dynamic feature vector representing consumption periodicity and capital flow patterns. The feature vector dimension is automatically aligned with the input layer of the credit scoring model; S5-2. The model decay monitor continuously calculates the relative change rate of the area under the ROC curve of the validation set, and when the model prediction performance decreases by more than a preset threshold, triggers an incremental training task containing samples from the latest time period, and the incremental training uses an elastic weight consolidation algorithm to prevent catastrophic forgetting; S5-3. The asynchronous update coordinator uses a historical gradient momentum buffer mechanism to optimize the parameter synchronization process, and maintains an independent version control queue for each participant, and determines the parameter fusion ratio by comparing the cosine similarity of the global model and the local model.

7. The cloud computing-based risk control credit monitoring method according to claim 1, characterized in that, The S6 specifically comprises the following steps: S6-1. The graph computing engine constructs a heterogeneous relationship graph containing enterprise shareholder chains, guarantee relationships, and actual controller paths, calculates node centrality scores using a random walk-based graph embedding algorithm, and a risk propagation model identifies hidden associated party credit risks by simulating capital flow; S6-2. The blockchain storage module organizes data operation records into a hash chain structure in chronological order, each block contains operation type, executor digital identity, and operation object fingerprint information, and storage data achieves distributed ledger consistency through Byzantine fault tolerance consensus algorithm; S6-3. The compliance verifier uses a non-interactive zero-knowledge proof protocol to verify cross-domain data transmission paths, the prover generates an evidence chain containing data hash values and routing nodes, and the verifier confirms that the transmission process has not been tampered with by an intermediate man through elliptic curve bilinear pair operations.

Images