Extensible aid decision-making method and system based on cross-border trade management

Abstract

The invention relates to an extensible aid decision-making method and system based on cross-border trade management, and belongs to the technical field of computer information processing. The method comprises the following steps: acquiring a cross-border trade auditing workflow map to bind an auxiliary decision-making unit, and when a business document is triggered, sending a business node decision-making request to the auxiliary decision-making unit according to packaged business data; constructing a multi-strategy auditing fusion model, obtaining an auxiliary decision-making unit list bound with the nodes, concurrently returning decision-making sub-results based on an asynchronous scheduling mechanism, and outputting a structured auditing evidence report; driving an auditing decision process according to a pre-configured branch routing rule, and when entering a manual auditing path, pushing the structured auditing evidence report as a task attachment to a manual auditor through a message array to generate decision feedback data; and regularly starting an optimization task to retrieve decision misjudgment cases in the review knowledge base, and updating the multi-strategy review fusion model.

Description

Technical Field

[0001] This invention belongs to the field of computer information processing technology, specifically relating to a scalable auxiliary decision-making method and system based on cross-border trade management. Background Technology

[0002] With the continuous expansion of cross-border e-commerce and international trade, cross-border trade management platforms need to process a large number of business documents involving customs declaration, price review, and compliance verification. Customs declaration price review is a crucial link in cross-border trade supervision, and its results directly affect customs clearance efficiency, tax compliance, and the level of trade risk control. However, existing cross-border trade review processes generally suffer from problems such as reliance on manual experience, rigid rules, and insufficient intelligence, making it difficult to adapt to the needs of rapid business growth and diversified trade scenarios.

[0003] In existing technologies, customs declaration price review is typically based on fixed rules or a single model. The review strategy is highly coupled with specific business processes, and once business rules or regulatory requirements change, the entire system process often needs to be adjusted, resulting in high system scalability and maintenance costs. Furthermore, the lack of effective coordination mechanisms between different review strategies makes it difficult to comprehensively utilize multiple methods such as price rules, historical data analysis, and risk models, thus limiting the accuracy and stability of the review results.

[0004] Existing cross-border trade review systems lack a unified mechanism for organizing and outputting evidence during the decision-making process. Review conclusions are often presented simply as pass or reject, failing to provide human reviewers with sufficient and explainable review evidence. This increases the workload of manual review and hinders subsequent auditing and accountability. In scenarios involving human intervention, the discrepancies between human review conclusions and automated system decisions are often not effectively recorded and utilized, preventing valuable review experience from being fed back into the system, leaving review strategies in a static state for extended periods.

[0005] The driving force for the review process is relatively simple, lacking a mechanism to dynamically adjust the review path based on risk level. This makes it difficult to achieve a reasonable separation between automatic and manual review, affecting overall review efficiency. Regarding model and rule updates, most systems still rely on manual maintenance, lacking a systematic closed-loop optimization mechanism, making it difficult to promptly identify misjudgment scenarios and continuously improve review capabilities.

[0006] Therefore, there is an urgent need for a cross-border trade audit support decision-making technology solution with scalability, interpretability, and self-evolution capabilities to support multi-strategy collaborative auditing, intelligent process-driven auditing, and continuous optimization based on feedback data, thereby improving the auditing efficiency and risk control level of the cross-border trade management system. Summary of the Invention

[0007] To address the aforementioned problems in the existing technology, this invention provides a scalable auxiliary decision-making method based on cross-border trade management. The objective of this invention can be achieved through the following technical solutions: S1: Obtain the cross-border trade review workflow diagram, bind the customs declaration price review node to the auxiliary decision unit, and store the binding relationship in the decision registry. When the business document is triggered, send a business node decision request to the auxiliary decision unit according to the encapsulated business data. S2: Construct a multi-strategy review fusion model based on the business node decision request, query the decision registry according to the customs declaration price review node identifier, obtain the list of auxiliary decision units bound to the node, and return the decision sub-results concurrently based on the asynchronous scheduling mechanism, and output a structured review evidence report; S3: The structured audit evidence report is sent back to the cross-border trade management platform. The audit decision process is driven to flow along the automatic or manual audit path according to the pre-configured branch routing rules. When the manual audit path is entered, a pending audit task is automatically generated, and the structured audit evidence report is pushed to the manual auditor as a task attachment through the message array to generate decision feedback data. S4: Construct a closed-loop feedback evolution mechanism, periodically initiate optimization tasks to retrieve and review decision misjudgment cases in the knowledge base, and automatically mine frequent risk patterns based on the decision feedback data to update the multi-strategy review fusion model.

[0008] Specifically, the method for obtaining the cross-border trade review workflow graph is as follows: collect the business node information corresponding to each review stage in the cross-border trade business, parse the parallel relationship between adjacent business nodes, and perform graph structure modeling on the business nodes and their connection relationships to generate a review workflow topology structure with business nodes as vertices and node flow relationships as directed edges, thereby obtaining the cross-border trade review workflow graph.

[0009] Specifically, the binding of the auxiliary decision-making unit includes a set of price reasonableness rules and a set of supplier blacklist checks; The price reasonableness rule set is as follows: Based on the commodity code and the country of trade, the benchmark value of the price range for the corresponding commodity is extracted. Based on the degree of deviation between the declared price in the business document and the benchmark value of the price range, and combined with the preset price fluctuation threshold, a price reasonableness score is generated. The supplier blacklist check set receives supplier identification information associated with business documents and matches and verifies it with a pre-maintained risk supplier blacklist database, generating a supplier risk level based on the supplier's historical violation records.

[0010] Specifically, the process of issuing the business node decision request is as follows: when the cross-border trade business document flows to the target business node in the review workflow graph, the customs declaration price review node event is triggered, and according to the node identifier of the target business node, the auxiliary decision unit identifier information bound to the business node is queried from the decision registry, and the business node decision request is sent to the corresponding auxiliary decision unit.

[0011] Specifically, the construction process of the multi-strategy audit fusion model is as follows: connecting with business node decision requests, extracting price features and supplier risk features from the business data carrier as multi-dimensional feature vectors, inputting the multi-dimensional feature vectors into the multi-strategy decision branches to statistically analyze audit decision sub-results, judging the confidence level of the results of the corresponding decision branches according to the preset audit strategy type, weighting and fusing the results of each strategy decision sub-result, outputting a unified audit decision result and corresponding risk explanation information, and constructing a multi-strategy audit fusion model.

[0012] Specifically, the method for obtaining the list of auxiliary decision-making units is as follows: Based on the business node identifier, a node-level index query is performed in the decision registry to obtain the associated auxiliary decision unit binding record. The validity of the auxiliary decision unit binding record is verified, and auxiliary decision units that are in a disabled state or do not meet the current business conditions are filtered out. A list of auxiliary decision units is obtained, including veto strategy, weighted voting strategy, and priority coverage strategy. In the veto strategy, if any decision unit's result is rejection, the entire decision is rejected. The weighted voting strategy calculates the weights based on the preset weights of each decision unit or the confidence level of the decision sub-results executed in this instance. The priority coverage strategy sets priorities for different decision units, with higher priority results covering lower priority results.

[0013] Specifically, the output process of the structured audit evidence report is as follows: based on the asynchronous scheduling mechanism, an independent asynchronous execution task is created for each auxiliary decision-making unit, and the corresponding business node decision requests are distributed and executed. The decision sub-results returned by each auxiliary decision-making unit are asynchronously received and cached in the result buffer. When all asynchronous execution tasks are completed, the decision sub-results in the result buffer are aggregated and the structured audit evidence report is output.

[0014] Specifically, the method for driving the review decision process is as follows: receiving the structured review evidence report, parsing the decision confidence index, matching the decision confidence index with the preset review strategy based on branch routing rules, and controlling the review decision process to continue to transfer along the automatic pass path when the decision confidence index is higher than the preset review pass threshold, and recording the corresponding review result log; when the decision confidence index is lower than the preset review pass threshold, controlling the review decision process to switch to the manual review path to verify the structured review evidence report.

[0015] Specifically, the method for generating the decision feedback data is as follows: after the review decision process enters the manual review path, a corresponding manual review task is automatically generated based on the current business document, and the structured review evidence report is associated with the manual review task as an attachment. The review operation result is associated with the corresponding structured review evidence report to generate decision feedback data containing the manual review conclusion and evidence tags.

[0016] Specifically, the method for constructing the closed-loop feedback evolution mechanism is as follows: periodically retrieve decision feedback data to identify misjudgment cases where the review results are inconsistent with the final decision, extract the frequent risk characteristics and strategy failure information from the misjudgment cases as a closed-loop analysis dataset to construct the closed-loop evolution mechanism.

[0017] Specifically, the method for updating the multi-strategy audit fusion model is as follows: when the closed-loop feedback evolution mechanism detects new decision feedback data, it parses the feedback data and historical audit records, verifies the audit decision indicators of the multi-strategy audit fusion model in actual business node decision-making, and redeploys the verified updated model to the auxiliary decision unit execution environment of the business node, replaces the original model in subsequent business node decision requests, records the model version number and adjustment parameters of the update process, and dynamically iterates and updates the multi-strategy audit fusion model.

[0018] Specifically, a scalable decision support system based on cross-border trade management includes: Workflow Decision Binding Module: Obtain the cross-border trade review workflow graph, bind auxiliary decision units to the customs declaration price review node, and store the binding relationship in the decision registry. When a business document is triggered, a business node decision request is sent to the auxiliary decision unit based on the encapsulated business data. Multi-strategy review fusion module: Constructs a multi-strategy review fusion model based on the business node decision request, queries the decision registry according to the customs declaration price review node identifier, obtains the list of auxiliary decision units bound to the node, and returns decision sub-results concurrently based on an asynchronous scheduling mechanism, and outputs a structured review evidence report; Audit-driven management module: The structured audit evidence report is sent back to the cross-border trade management platform. The audit decision process is driven to flow along the automatic audit or manual audit path according to the pre-configured branch routing rules. When the manual audit path is entered, a pending audit task is automatically generated, and the structured audit evidence report is pushed to the manual auditor as a task attachment through the message array to generate decision feedback data. Closed-loop feedback evolution module: Construct a closed-loop feedback evolution mechanism, periodically initiate optimization tasks to retrieve and review decision misjudgment cases in the knowledge base, and automatically mine frequent risk patterns based on the decision feedback data to update the multi-strategy review fusion model.

[0019] The beneficial effects of this invention are as follows: Compared with existing technologies, the scalable auxiliary decision-making method based on cross-border trade management proposed in this invention decouples audit strategies from business processes by introducing auxiliary decision-making units and a multi-strategy audit fusion mechanism into the cross-border trade audit workflow, significantly improving the system's scalability and flexibility. By using a decision registry to uniformly manage business nodes and auxiliary decision-making units, different audit strategies can be dynamically bound as needed, facilitating the addition, removal, and optimization of strategies without affecting existing business processes.

[0020] By employing a multi-strategy audit fusion model to concurrently process and weightedly fuse decision sub-results from different auxiliary decision-making units, the model comprehensively utilizes various audit methods such as rule validation, statistical analysis, and model inference to improve the accuracy and stability of customs declaration price audit results. Simultaneously, the structured audit evidence report output provides clear and interpretable decision-making basis for both automated and manual audits, reducing the understanding costs and decision-making risks for manual auditors.

[0021] Based on pre-configured branch routing rules, the review process is dynamically driven to switch between automatic and manual review paths according to the comprehensive risk level, achieving reasonable allocation of review resources and effectively improving overall review efficiency. In the manual review scenario, the automatic generation of to-do tasks and message array push mechanism enable efficient distribution of review tasks, and the manual review results are used to form decision feedback data.

[0022] A closed-loop feedback evolution mechanism has been constructed, which can automatically mine frequent risk patterns based on historical misjudgment cases and decision feedback data, and continuously update and optimize the multi-strategy audit fusion model, enabling the audit strategy to have self-learning and self-evolution capabilities, thereby continuously improving the risk identification capability and long-term operational effectiveness of the cross-border trade audit system. Attached Figure Description

[0023] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0024] Figure 1 This is a schematic diagram of the framework of a scalable auxiliary decision-making method and system based on cross-border trade management according to the present invention.

[0025] Figure 2 This is a schematic diagram illustrating the verification of an scalable auxiliary decision-making method and system based on cross-border trade management according to the present invention. Detailed Implementation

[0026] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0027] Please see Figure 1 A scalable decision support method based on cross-border trade management: S1: Obtain the cross-border trade review workflow diagram, bind the customs declaration price review node to the auxiliary decision unit, and store the binding relationship in the decision registry. When the business document is triggered, send a business node decision request to the auxiliary decision unit according to the encapsulated business data. S2: Construct a multi-strategy review fusion model based on the business node decision request, query the decision registry according to the customs declaration price review node identifier, obtain the list of auxiliary decision units bound to the node, and return the decision sub-results concurrently based on the asynchronous scheduling mechanism, and output a structured review evidence report; S3: The structured audit evidence report is sent back to the cross-border trade management platform. The audit decision process is driven to flow along the automatic or manual audit path according to the pre-configured branch routing rules. When the manual audit path is entered, a pending audit task is automatically generated, and the structured audit evidence report is pushed to the manual auditor as a task attachment through the message array to generate decision feedback data. S4: Construct a closed-loop feedback evolution mechanism, periodically initiate optimization tasks to retrieve and review decision misjudgment cases in the knowledge base, and automatically mine frequent risk patterns based on the decision feedback data to update the multi-strategy review fusion model.

[0028] In this embodiment, the method for obtaining the cross-border trade review workflow graph is as follows: collect the business node information corresponding to each review stage in the cross-border trade business, parse the parallel relationship between adjacent business nodes, and perform graph structure modeling on the business nodes and their connection relationships to generate a review workflow topology structure with business nodes as vertices and node flow relationships as directed edges, thereby obtaining the cross-border trade review workflow graph.

[0029] In this embodiment, the binding of the auxiliary decision-making unit includes a price reasonableness rule set and a supplier blacklist check set; The price reasonableness rule set is as follows: Based on the commodity code and the country of trade, the benchmark value of the price range for the corresponding commodity is extracted. Based on the degree of deviation between the declared price in the business document and the benchmark value of the price range, and combined with the preset price fluctuation threshold, a price reasonableness score is generated. The supplier blacklist check set receives supplier identification information associated with business documents and matches and verifies it with a pre-maintained risk supplier blacklist database, generating a supplier risk level based on the supplier's historical violation records.

[0030] In this embodiment, the process of issuing the business node decision request is as follows: when the cross-border trade business document flows to the target business node in the review workflow graph, the customs declaration price review node event is triggered, and according to the node identifier of the target business node, the auxiliary decision unit identifier information bound to the business node is queried from the decision registry, and the business node decision request is sent to the corresponding auxiliary decision unit.

[0031] In this embodiment, the system development option is: The workflow engine chosen is the open-source project Camunda.

[0032] The decision-making center is developed using the Spring Boot framework and provides a RESTful API.

[0033] The decision unit development framework provides a Java / Python SDK to standardize input and output data formats.

[0034] Redis is used to cache decision unit registration information and hot data, while MySQL is used to store knowledge base and configuration information.

[0035] Example of decision-making unit development (product classification prompt model unit): The developers used Python to train a BERT-based text classification model to predict possible customs commodity codes based on product descriptions.

[0036] According to the SDK specification, the model is encapsulated as a service and the execute(Context ctx) method of the DecisionUnit interface is implemented.

[0037] Register the cell in the management console, name it CommodityCodeHintModel, and configure its metadata (version, author, input / output description).

[0038] Business process configuration example: At the "Customs Declaration Entry Completed" review node, the administrator bound three decision-making units by dragging and dropping: BasicDataValidationRule CommodityCodeHintModel (Product Classification Hint Model) EnterpriseCreditCheckService The configuration fusion strategy is as follows: the basic verification rules have veto power; the classification prompts and credit query results are only used as alarm information for manual reference.

[0039] Runtime example: like Figure 2 A company submits a customs declaration. Upon reaching this node in the process, the decision-making center concurrently invokes three units.

[0040] Unit 1 passed; Unit 2 returned the following message: "Based on the description 'carbon steel bolts', the predicted code may be 7318.15, which does not match your declared 7318.12, with a confidence level of 85%"; Unit 3 returned: "The company's credit rating is B, and it has one minor violation in its history."

[0041] Merger generation suggestion: "Basic verification passed. Hint:" There appears to be an error in the product classification; a review is recommended. The applicant company's credit rating is average. The workflow engine then generates a "pending review" task and sends the aforementioned explicit prompts to the customs reviewer, enabling them to quickly focus on the risk points.

[0042] This invention is not only applicable to cross-border trade, but can also be extended to any intelligent decision-making scenario involving complex, multi-rule, and evolvable business processes, such as bank credit approval, insurance claims, and content security review.

[0043] In this embodiment, the construction process of the multi-strategy audit fusion model is as follows: connecting to the business node decision request, extracting the price features and supplier risk features from the business data carrier as multi-dimensional feature vectors, inputting the multi-dimensional feature vectors into the multi-strategy decision branches to statistically analyze the audit decision sub-results, judging the confidence level of the results of the corresponding decision branches according to the preset audit strategy type, weighting and fusing the results of each strategy decision sub-result, outputting a unified audit decision result and corresponding risk explanation information, and constructing the multi-strategy audit fusion model.

[0044] In this embodiment, the method for obtaining the list of auxiliary decision-making units is as follows: Based on the business node identifier, a node-level index query is performed in the decision registry to obtain the associated auxiliary decision unit binding record. The validity of the auxiliary decision unit binding record is verified, and auxiliary decision units that are in a disabled state or do not meet the current business conditions are filtered out. A list of auxiliary decision units is obtained, including veto strategy, weighted voting strategy, and priority coverage strategy. In the veto strategy, if any decision unit's result is rejection, the entire decision is rejected. The weighted voting strategy calculates the weights based on the preset weights of each decision unit or the confidence level of the decision sub-results executed in this instance. The priority coverage strategy sets priorities for different decision units, with higher priority results covering lower priority results.

[0045] In this embodiment, the output process of the structured audit evidence report is as follows: based on the asynchronous scheduling mechanism, an independent asynchronous execution task is created for each auxiliary decision-making unit, and the corresponding business node decision requests are distributed and executed. The decision sub-results returned by each auxiliary decision-making unit are asynchronously received and cached in the result buffer. When all asynchronous execution tasks are completed, the decision sub-results in the result buffer are aggregated and the structured audit evidence report is output.

[0046] In this embodiment, the method for driving the review decision process is as follows: receiving the structured review evidence report and parsing the decision confidence index; matching the decision confidence index with a preset review strategy based on branch routing rules; when the decision confidence index is higher than a preset review pass threshold, controlling the review decision process to continue to transfer along the automatic pass path and recording the corresponding review result log; when the decision confidence index is lower than the preset review pass threshold, controlling the review decision process to switch to the manual review path to verify the structured review evidence report.

[0047] In this embodiment, the decision registry adopts a relational table structure, and multiple auxiliary decision units can be bound to the same node, as shown in the following example: RULE_PRICE_RANGE (price range rule); RULE_SUPPLIER_BLACKLIST (Supplier blacklist); MODEL_PRICE_PREDICT (Price Prediction Model); When a business document triggers the customs declaration price review node, a business node decision request is generated. The multi-strategy review fusion model includes three strategy sub-modules: 1. Rules and Strategies Submodule; Based on historical customs declaration databases, the price range for similar commodities is statistically analyzed. The historical mean μ is 100, and the historical standard deviation σ is 10. The reasonable price range is defined as follows: , The current quoted price is 95 USD. Calculate the standardized deviation value: , Rule-based strategies output risk scores: .

[0048] 2. Statistical Analysis Strategy Submodule; Based on historical anomaly rate statistics, calculate the anomaly probability of the current bid price, and substitute it into z=0.5 to set the anomaly probability function: , The statistical strategy risk score is defined as follows: .

[0049] 3. Model Inference Strategy Submodule; Predicting fair prices using a linear regression model: , Where x1 is the historical average price of the commodity, x2 is the exchange rate factor, and the weights are w1=0.9, w2=0.1, and b=0.

[0050] The predicted price is: , The prediction bias is: , The risk score of the model strategy is: .

[0051] The system adopts an asynchronous scheduling mechanism: an asynchronous task is generated for each auxiliary decision-making unit; the timeout period is set to 300ms; the maximum concurrency is 8; the convergence condition is that all tasks are completed or the timeout task is completed, and the default risk value of 0.5 is automatically returned.

[0052] Multi-strategy review fusion algorithm fusion formula: , , Among them, R final ≥0.8 Automatic approval; 0.5≤R final <0.8 manual review; R final <0.5 forced manual labor, .

[0053] In a closed-loop feedback evolution mechanism, if: , Then it is marked as a misjudged sample.

[0054] Gradient correction is used: , The learning rate is η=0.05. After updating, the weights are renormalized to generate a new model version.

[0055] In this embodiment, the method for generating the decision feedback data is as follows: after the review decision process enters the manual review path, a corresponding manual review task is automatically generated based on the current business document, and the structured review evidence report is associated with the manual review task as a task attachment. The review operation result is associated with the corresponding structured review evidence report to generate decision feedback data containing the manual review conclusion and evidence tags.

[0056] In this embodiment, the method for constructing the closed-loop feedback evolution mechanism is as follows: periodically retrieve decision feedback data to identify misjudgment cases where the review results are inconsistent with the final decision, and extract the frequent risk characteristics and strategy failure information from the misjudgment cases as a closed-loop analysis dataset to construct the closed-loop evolution mechanism.

[0057] In this embodiment, the method for updating the multi-strategy review fusion model is as follows: when the closed-loop feedback evolution mechanism detects new decision feedback data, it parses the feedback data and historical review records, verifies the review decision indicators of the multi-strategy review fusion model in actual business node decision-making, and redeploys the verified updated model to the auxiliary decision unit execution environment of the business node, replaces the original model in subsequent business node decision requests, records the model version number and adjustment parameters of the update process, and dynamically iterates and updates the multi-strategy review fusion model.

[0058] This invention also provides a scalable auxiliary decision-making system based on cross-border trade management, specifically including: Workflow Decision Binding Module: Obtain the cross-border trade review workflow graph, bind auxiliary decision units to the customs declaration price review node, and store the binding relationship in the decision registry. When a business document is triggered, a business node decision request is sent to the auxiliary decision unit based on the encapsulated business data. Multi-strategy review fusion module: Constructs a multi-strategy review fusion model based on the business node decision request, queries the decision registry according to the customs declaration price review node identifier, obtains the list of auxiliary decision units bound to the node, and returns decision sub-results concurrently based on an asynchronous scheduling mechanism, and outputs a structured review evidence report; Audit-driven management module: The structured audit evidence report is sent back to the cross-border trade management platform. The audit decision process is driven to flow along the automatic audit or manual audit path according to the pre-configured branch routing rules. When the manual audit path is entered, a pending audit task is automatically generated, and the structured audit evidence report is pushed to the manual auditor as a task attachment through the message array to generate decision feedback data. Closed-loop feedback evolution module: Construct a closed-loop feedback evolution mechanism, periodically initiate optimization tasks to retrieve and review decision misjudgment cases in the knowledge base, and automatically mine frequent risk patterns based on the decision feedback data to update the multi-strategy review fusion model.

[0059] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A scalable decision support method based on cross-border trade management, characterized in that, include: S1: Obtain the cross-border trade review workflow diagram, bind the customs declaration price review node to the auxiliary decision unit, and store the binding relationship in the decision registry. When the business document is triggered, send a business node decision request to the auxiliary decision unit according to the encapsulated business data. S2: Construct a multi-strategy review fusion model based on the business node decision request, query the decision registry according to the customs declaration price review node identifier, obtain the list of auxiliary decision units bound to the node, and return the decision sub-results concurrently based on the asynchronous scheduling mechanism, and output a structured review evidence report; S3: The structured audit evidence report is sent back to the cross-border trade management platform. The audit decision process is driven to flow along the automatic or manual audit path according to the pre-configured branch routing rules. When the manual audit path is entered, a pending audit task is automatically generated, and the structured audit evidence report is pushed to the manual auditor as a task attachment through the message array to generate decision feedback data. S4: Construct a closed-loop feedback evolution mechanism, periodically initiate optimization tasks to retrieve and review decision misjudgment cases in the knowledge base, and automatically mine frequent risk patterns based on the decision feedback data to update the multi-strategy review fusion model.

2. The method according to claim 1, characterized in that, The method for obtaining the cross-border trade review workflow graph is as follows: collect the business node information corresponding to each review stage in the cross-border trade business, parse the parallel relationship between adjacent business nodes, and perform graph structure modeling on the business nodes and their connection relationships to generate a review workflow topology structure with business nodes as vertices and node flow relationships as directed edges, thereby obtaining the cross-border trade review workflow graph.

3. The method according to claim 1, characterized in that, The binding of the auxiliary decision-making unit includes a set of price reasonableness rules and a set of supplier blacklist checks; The price reasonableness rule set is as follows: Based on the commodity code and the country of trade, the benchmark value of the price range for the corresponding commodity is extracted. Based on the degree of deviation between the declared price in the business document and the benchmark value of the price range, and combined with the preset price fluctuation threshold, a price reasonableness score is generated. The supplier blacklist check set receives supplier identification information associated with business documents and matches and verifies it with a pre-maintained risk supplier blacklist database, generating a supplier risk level based on the supplier's historical violation records.

4. The method according to claim 1, characterized in that, The process of issuing the business node decision request is as follows: when the cross-border trade business document flows to the target business node in the review workflow graph, the customs declaration price review node event is triggered, and according to the node identifier of the target business node, the auxiliary decision unit identifier information bound to the business node is queried from the decision registry, and the business node decision request is sent to the corresponding auxiliary decision unit.

5. The method according to claim 2, characterized in that, The construction process of the multi-strategy audit fusion model is as follows: connect to the business node decision request, extract the price features and supplier risk features from the business data carrier as multi-dimensional feature vectors, input the multi-dimensional feature vectors into the multi-strategy decision branches to statistically analyze the audit decision sub-results, and judge the confidence level of the results of the corresponding decision branches according to the preset audit strategy type. The results of each strategy decision sub-result are weighted and fused to output a unified audit decision result and corresponding risk explanation information, thus constructing the multi-strategy audit fusion model.

6. The method according to claim 5, characterized in that, The method for obtaining the list of auxiliary decision-making units is as follows: Based on the business node identifier, a node-level index query is performed in the decision registry to obtain the associated auxiliary decision unit binding record. The validity of the auxiliary decision unit binding record is verified, and auxiliary decision units that are in a disabled state or do not meet the current business conditions are filtered out. A list of auxiliary decision units is obtained, including veto strategy, weighted voting strategy, and priority coverage strategy. In the veto strategy, if any decision unit's result is rejection, the entire decision is rejected. The weighted voting strategy calculates the weights based on the preset weights of each decision unit or the confidence level of the decision sub-results executed in this instance. The priority coverage strategy sets priorities for different decision units, with higher priority results covering lower priority results.

7. The method according to claim 4, characterized in that, The output process of the structured audit evidence report is as follows: Based on the asynchronous scheduling mechanism, an independent asynchronous execution task is created for each auxiliary decision-making unit, and the corresponding business node decision requests are distributed and executed. The decision sub-results returned by each auxiliary decision-making unit are asynchronously received and cached in the result buffer. When all asynchronous execution tasks are completed, the decision sub-results in the result buffer are aggregated and the structured audit evidence report is output.

8. The method according to claim 2, characterized in that, The method for driving the review decision process is as follows: receiving the structured review evidence report, parsing the decision confidence index, matching the decision confidence index with the preset review strategy based on the branch routing rules, and controlling the review decision process to continue to transfer along the automatic pass path when the decision confidence index is higher than the preset review pass threshold, and recording the corresponding review result log; when the decision confidence index is lower than the preset review pass threshold, controlling the review decision process to switch to the manual review path to verify the structured review evidence report.

9. The method according to claim 4, characterized in that, The method for generating the decision feedback data is as follows: after the review decision process enters the manual review path, a corresponding manual review task is automatically generated based on the current business document, and the structured review evidence report is associated with the manual review task as an attachment. The review operation result is associated with the corresponding structured review evidence report to generate decision feedback data containing the manual review conclusion and evidence tags.

10. The method according to claim 2, characterized in that, The method for constructing the closed-loop feedback evolution mechanism is as follows: periodically retrieve decision feedback data to identify misjudgment cases where the review results are inconsistent with the final decision, extract the frequent risk characteristics and strategy failure information from the misjudgment cases as a closed-loop analysis dataset to construct the closed-loop evolution mechanism.

11. The method according to claim 7, characterized in that, The method for updating the multi-strategy audit fusion model is as follows: when the closed-loop feedback evolution mechanism detects new decision feedback data, it parses the feedback data and historical audit records, verifies the audit decision indicators of the multi-strategy audit fusion model in actual business node decision-making, and redeploys the verified updated model to the auxiliary decision unit execution environment of the business node. In subsequent business node decision requests, it replaces the original model, records the model version number and adjustment parameters of the update process, and dynamically iterates and updates the multi-strategy audit fusion model.

12. A scalable decision support system based on cross-border trade management, used to execute the method as described in any one of claims 1-11, characterized in that, include: Workflow Decision Binding Module: Obtain the cross-border trade review workflow graph, bind auxiliary decision units to the customs declaration price review node, and store the binding relationship in the decision registry. When a business document is triggered, a business node decision request is sent to the auxiliary decision unit based on the encapsulated business data. Multi-strategy review fusion module: Constructs a multi-strategy review fusion model based on the business node decision request, queries the decision registry according to the customs declaration price review node identifier, obtains the list of auxiliary decision units bound to the node, and returns decision sub-results concurrently based on an asynchronous scheduling mechanism, and outputs a structured review evidence report; Audit-driven management module: The structured audit evidence report is sent back to the cross-border trade management platform. The audit decision process is driven to flow along the automatic audit or manual audit path according to the pre-configured branch routing rules. When the manual audit path is entered, a pending audit task is automatically generated, and the structured audit evidence report is pushed to the manual auditor as a task attachment through the message array to generate decision feedback data. Closed-loop feedback evolution module: Construct a closed-loop feedback evolution mechanism, periodically initiate optimization tasks to retrieve and review decision misjudgment cases in the knowledge base, and automatically mine frequent risk patterns based on the decision feedback data to update the multi-strategy review fusion model.

Images