Method for checking consistency of data of old-for-new commodity based on GS1 coding rule engine

Abstract

The application discloses a method for checking data consistency of old-for-new goods based on a GS1 coding rule engine, which comprises the following steps: obtaining a multi-source standardized record set by performing semantic mapping and standardization processing on multi-source heterogeneous data, calculating multi-source consistency measurement, performing cross-source data consistency checking, constructing a multi-source space-time correlation graph, calculating the space-time correlation strength of each edge in the graph and the local subgraph anomaly score centered on the current order, extracting real-time load and merchant information of the checking system, inputting the local subgraph anomaly score into a checking depth decision function, obtaining the checking depth level and checking sampling frequency of the current order, performing differential checking according to the checking depth level of the current order and outputting a checking conclusion, and optimizing the checking depth decision function according to the checking effect. The method can realize high-risk deep checking, low-risk quick release and automatic degradation in peak periods, and improves the data consistency checking accuracy and subsidy fund safety of the old-for-new business.

Description

Technical Field

[0001] This invention relates to the field of data verification technology, and in particular to a method for verifying the consistency of trade-in product data based on the GS1 coding rule engine. Background Technology

[0002] The trade-in program for consumer goods involves collaboration among multiple entities, including sales, recycling, dismantling, subsidies, and merchant registration. Key data such as product identification, transaction vouchers, consumer information, the flow of old devices, and merchant qualifications are scattered across heterogeneous platforms. Cross-platform data consistency verification is a core step in ensuring the safety of subsidy funds and preventing fraudulent subsidy activities.

[0003] Existing technologies mostly employ single-point rule verification or fixed-ratio sampling, lacking cross-system semantic alignment capabilities, struggling to handle heterogeneous data mapping issues, and exhibiting rigid verification strategies that cannot dynamically adjust depth based on transaction risk and system load. This leads to over-verification of low-risk orders and under-verification of high-risk orders, making it difficult to identify cross-order fraud groups. Therefore, this invention proposes a data consistency verification method for trade-in products based on the GS1 coding rule engine. Through semantic standardization, spatiotemporal correlation graph anomaly scoring, and load-adaptive depth decision-making, it achieves accurate alignment of multi-source data and optimized allocation of verification resources, improving the risk control efficiency and fund security of trade-in businesses. Summary of the Invention

[0004] The purpose of this invention is to provide a method for verifying the consistency of trade-in product data based on the GS1 coding rule engine.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: By accessing the multi-source heterogeneous data of the old-for-new business through standard APIs, the fields of each source data are semantically mapped and standardized through the GS1 semantic alignment tree to obtain a multi-source standardized record set. Calculate a multi-source consistency metric based on a multi-source standardized record set, and perform cross-source data consistency verification. A multi-source spatiotemporal correlation graph is constructed based on the GS1 encoding of the multi-source standardized record set as the correlation link. The spatiotemporal correlation strength of each edge in the graph and the anomaly score of the local subgraph centered on the current order are calculated. Extract the real-time load and merchant information of the verification system, and combine the local subgraph anomaly score with the input verification depth decision function to obtain the verification depth level and verification sampling frequency of the current order; Perform differentiated verification based on the current order's verification depth level and output verification conclusions; optimize the verification depth decision function based on the verification results. The multi-source heterogeneous data includes product identification information, transaction voucher information, consumer information, used device information, and merchant information; the multi-source heterogeneous data comes from sales platforms, recycling platforms, dismantling companies, subsidy platforms, and merchant registration systems; The differentiated verification includes single-chain lightweight verification processing, related transaction chain verification, and full-scale in-depth review.

[0006] Furthermore, the method for obtaining the multi-source normalized recordset includes: Construct a GS1 five-source semantic alignment tree, with the GS1 application identifier as the root node, and expand downwards to data domain nodes, format nodes, and value domain constraint nodes. Calculate the mapping confidence of each source data field to all leaf nodes of the GS1 semantic tree. When the mapping confidence is greater than the high confidence threshold, the corresponding source data field is directly mapped to the corresponding leaf node. When the mapping confidence is less than the low confidence threshold, the mapping of the corresponding source data field is rejected. Otherwise, the corresponding leaf node is used as a candidate leaf node for disambiguation processing. The disambiguation process calculates the disambiguation score of each candidate leaf node using a hierarchical disambiguation function, and selects the node with the highest disambiguation score as the final mapping target. Complete the semantic mapping of all source data fields to obtain a multi-source standardized record set; The formula for calculating the mapping confidence level is: ; in For fields To the leaf node The mapping confidence, , , For mapping weights, For data source Middle field The tag, Leaf nodes in the GS1 semantic tree The tag, For the similarity of field label characters, For fields Value range distribution, For nodes The standard range For the similarity of value range distribution, For fields Contextual information, The semantic node conditional probability is given by the context. The disambiguation score is calculated as follows: ; in For nodes The disambiguation score, , The disambiguation coefficient is... For nodes Depth in the semantic tree For nodes The number of competing nodes at the same level.

[0007] Furthermore, the method for performing cross-source data consistency verification includes: A multi-source consistency metric is calculated based on a multi-source standardized record set. Cross-source data consistency is then checked based on the multi-source consistency metric. If the multi-source consistency metric is lower than a preset threshold, a cross-source data inconsistency alarm is triggered and the current order verification process is terminated. Otherwise, the next step of verification is performed. The multi-source consistency metric is used to evaluate the degree of consistency of multi-source data at key semantic nodes. It is determined by calculating the node consistency of each data source at the corresponding node using the Gaussian radial basis kernel function and weighting the results.

[0008] Furthermore, the method for calculating the spatiotemporal correlation strength of each edge in the graph and the anomaly score of the local subgraph centered on the current order includes: The entities in the multi-source standardized record set are abstracted into graph nodes to form a node set. The graph edges are constructed based on the business relationships of the old-for-new business to form an edge set. The spatiotemporal correlation strength of each edge in the graph is calculated based on the time difference, spatial distance, and GS1 coding sharing level between each graph node. The spatiotemporal correlation strength is negatively correlated with the time difference and spatial distance, and positively correlated with the GS1 coding sharing level. The GS1 coding sharing level is determined by the GS1 coding sharing level indicator function. The GS1 code sharing level indicator function is characterized by the following: it takes 1 when two graph nodes share the same AI 21 single item serial number, 0.8 when two graph nodes share the same AI 10 production batch number, 0.5 when two graph nodes share the same GTIN product category code but have different single item serial numbers, and 0 when two graph nodes have no GS1 code association. Based on the historical violation rate, related order density, amount deviation and time concentration of each graph node, the corresponding risk feature scalar is determined. The spatiotemporal correlation strength is weighted and aggregated by the risk feature scalar and edge relationship to obtain the local subgraph anomaly score centered on the current order.

[0009] Furthermore, the method for obtaining the verification depth level and verification sampling frequency of the current order includes: Extract the real-time load of the verification system and merchant information, and calculate the real-time load coefficient of the system and the merchant risk profile score respectively; the real-time load coefficient of the system is the ratio of the current processing queue length to the maximum throughput of the system; the merchant risk profile score is calculated based on the historical compliance rate, complaint rate and return rate. Input the system's real-time load coefficient, merchant risk profile score, and local subgraph anomaly score into the verification depth decision function to obtain the verification depth level and verification sampling frequency for the current order. The expression is: ; ; in For orders The level of verification depth, For orders Local subgraph centered Abnormal scores, This is the threshold for the anomaly scoring benchmark. For merchants Risk profile scoring for Real-time load factor of the system For load sensitivity coefficient, This is a truncation function that cuts the output value to the interval between 1 and 3. For orders The verification sampling frequency, The maximum sampling frequency supported by the system. The sampling sensitivity coefficient, This is the threshold for abnormal scoring risk. It is a sigmoid activation function.

[0010] Furthermore, the single-chain lightweight verification process only performs basic consistency comparison processing on the current transaction order in the areas of product identity verification, transaction certificate verification, consumer information verification, old machine information verification, and merchant registration verification.

[0011] Furthermore, the related transaction chain verification also performs single-chain lightweight verification processing on the current order and neighboring orders that are related to the current order.

[0012] Furthermore, the full-volume in-depth review performs full consistency verification on all types of current transaction orders, including product identity verification, transaction voucher verification, consumer information verification, used device information verification, and merchant registration verification. After reviewing historical orders of the same type, the results are output to the manual final review queue. The historical order backtracking and verification involves retrieving historical orders that have the same merchant identifier, the same consumer identifier, or the same used product category identifier as the current order, and identifying abnormal historical orders across time dimensions. The historical similar order retrospective review adopts a strategy based on marginal risk and return, and calculates the marginal risk and return to iterate and solve for the optimal retrospective parameters to determine the retrospective range; the termination condition of the iterative solution is that the cumulative review cost reaches the daily manual review budget limit or the marginal risk and return is lower than the preset termination threshold.

[0013] Furthermore, the method for optimizing the verification depth decision function based on the verification effect includes: Based on the verification results, a strategy effectiveness evaluation function is constructed. With the goal of maximizing the strategy effectiveness evaluation function, the dynamic parameters in the verification depth decision function are optimized online using a zero-order optimization method. The dynamic parameters include the anomaly scoring benchmark threshold, the load sensitivity coefficient, the sampling sensitivity coefficient, and the anomaly scoring risk threshold.

[0014] The beneficial effects of this invention are: This invention is a data consistency verification method for trade-in products based on the GS1 coding rule engine. Compared with existing technologies, this invention has the following technical advantages: This invention achieves standardized mapping of multi-source heterogeneous data through GS1 semantic alignment tree, eliminating the semantic gap across systems; This invention quantifies and evaluates cross-platform data consistency through a multi-source consistency metric; This invention expands the risk control dimension from a single transaction to the entire transaction network by constructing a spatiotemporal correlation graph to calculate anomaly scores for local subgraphs, thereby identifying cross-order group fraud. This invention dynamically outputs the verification depth level and sampling frequency by integrating anomaly scoring and system load into a verification depth decision function, thereby achieving high-risk deep verification, low-risk rapid release, and automatic downgrading during peak periods. This invention continuously optimizes the decision function through verification effect feedback, enabling the system to have self-evolution capabilities and improving the accuracy of data consistency verification and the security of subsidy funds in the old-for-new business. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the steps of the data consistency verification method for trade-in products based on the GS1 encoding rule engine of this invention. Detailed Implementation

[0016] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0017] The present invention provides a data consistency verification method for trade-in products based on the GS1 coding rule engine, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: By accessing the multi-source heterogeneous data of the old-for-new business through standard APIs, the fields of each source data are semantically mapped and standardized through the GS1 semantic alignment tree to obtain a multi-source standardized record set. Calculate a multi-source consistency metric based on a multi-source standardized record set, and perform cross-source data consistency verification. A multi-source spatiotemporal correlation graph is constructed based on the GS1 encoding of the multi-source standardized record set as the correlation link. The spatiotemporal correlation strength of each edge in the graph and the anomaly score of the local subgraph centered on the current order are calculated. Extract the real-time load and merchant information of the verification system, and combine the local subgraph anomaly score with the input verification depth decision function to obtain the verification depth level and verification sampling frequency of the current order; Perform differentiated verification based on the current order's verification depth level and output verification conclusions; optimize the verification depth decision function based on the verification results. The multi-source heterogeneous data includes product identification information, transaction voucher information, consumer information, used device information, and merchant information; the multi-source heterogeneous data comes from sales platforms, recycling platforms, dismantling companies, subsidy platforms, and merchant registration systems; The differentiated verification includes single-chain lightweight verification processing, related transaction chain verification, and full-scale in-depth review.

[0018] In this embodiment, the method for obtaining a multi-source normalized recordset includes: Construct a GS1 five-source semantic alignment tree, with the GS1 application identifier as the root node, and expand downwards to data domain nodes, format nodes, and value domain constraint nodes. Calculate the mapping confidence of each source data field to all leaf nodes of the GS1 semantic tree. When the mapping confidence is greater than the high confidence threshold, the corresponding source data field is directly mapped to the corresponding leaf node. When the mapping confidence is less than the low confidence threshold, the mapping of the corresponding source data field is rejected. Otherwise, the corresponding leaf node is used as a candidate leaf node for disambiguation processing. The disambiguation process calculates the disambiguation score of each candidate leaf node using a hierarchical disambiguation function, and selects the node with the highest disambiguation score as the final mapping target. Complete the semantic mapping of all source data fields to obtain a multi-source standardized record set; The formula for calculating the mapping confidence level is: ; in For fields To the leaf node The mapping confidence, , , For mapping weights, For data source Middle field The tag, Leaf nodes in the GS1 semantic tree The tag, For the similarity of field label characters, For fields Value range distribution, For nodes The standard range For the similarity of value range distribution, For fields Contextual information, The semantic node conditional probability is given by the context. The disambiguation score is calculated as follows: ; in For nodes The disambiguation score, , The disambiguation coefficient is... For nodes Depth in the semantic tree For nodes The number of competing nodes at the same level; In actual evaluation, the GS1 semantic tree takes the Application Identifier (AI) defined by the GS1 General Specification as the root node, and extends downwards to data field nodes (such as the AI ​​21 serial number field containing child nodes such as IMEI, SN, MEID, etc.), format nodes (pure numbers and mixed alphanumeric), and value field constraint nodes. The similarity of the field label characters is calculated using normalized edit distance, the similarity of the value range distribution is measured by histogram intersection or KL divergence, and the conditional probability of the semantic node is output by the pre-trained GS1 semantic association model, with a mapping weight sum of 1, and is determined by the verification scenario through optimization using historical labeled data. When the mapping confidence of all target mapping leaf nodes for fields in the same data source is greater than the high confidence threshold, all such leaf nodes are treated as candidate leaf nodes and the disambiguation process is performed. If only one target mapping leaf node has a mapping confidence greater than the high confidence threshold, the corresponding field in the data source is directly mapped to that node. If no target mapping leaf nodes have mapping confidence greater than the high confidence threshold, but multiple target mapping leaf nodes have mapping confidence between the low and high confidence thresholds, the relevant target mapping leaf nodes are treated as candidate leaf nodes and the disambiguation process is performed. The multi-source standardized record set is formed by transforming multi-source heterogeneous data into a standardized value vector on the GS1 semantic tree key node set (such as GTIN, serial number, price, energy efficiency level) after semantic mapping.

[0019] In this embodiment, the method for performing cross-source data consistency verification includes: A multi-source consistency metric is calculated based on a multi-source standardized record set. Cross-source data consistency is then checked based on the multi-source consistency metric. If the multi-source consistency metric is lower than a preset threshold, a cross-source data inconsistency alarm is triggered and the current order verification process is terminated. Otherwise, the next step of verification is performed. The multi-source consistency metric is used to evaluate the degree of consistency of multi-source data at key semantic nodes. It is determined by calculating the node consistency of each data source at the corresponding node through the Gaussian radial basis kernel function and weighting the results. In practical evaluation, the multi-source consistency metric is calculated as follows: ; ; in For multi-source consistency measurement, For a multi-source standardized record set, For data source collection, Indexing the data source, This is the set of key semantic nodes in the GS1 semantic tree. For nodes Business weight, For data source At the node The standardized value vector on, For consistency kernel function, For nodes Historically consistent sample centers For adaptive bandwidth, For data source Historical error penalty coefficient; The historical error penalty coefficient is used to adaptively correct for the reliability differences of different data sources. The higher the reliability of the data source, the smaller the historical error penalty coefficient, and the larger the corresponding consistency kernel function output.

[0020] In this embodiment, the method for calculating the spatiotemporal correlation strength of each edge in the graph and the anomaly score of the local subgraph centered on the current order includes: The entities in the multi-source standardized record set are abstracted into graph nodes to form a node set. The graph edges are constructed based on the business relationships of the old-for-new business to form an edge set. The spatiotemporal correlation strength of each edge in the graph is calculated based on the time difference, spatial distance, and GS1 coding sharing level between each graph node. The spatiotemporal correlation strength is negatively correlated with the time difference and spatial distance, and positively correlated with the GS1 coding sharing level. The GS1 coding sharing level is determined by the GS1 coding sharing level indicator function. The GS1 code sharing level indicator function is characterized by the following: it takes 1 when two graph nodes share the same AI 21 single item serial number, 0.8 when two graph nodes share the same AI 10 production batch number, 0.5 when two graph nodes share the same GTIN product category code but have different single item serial numbers, and 0 when two graph nodes have no GS1 code association. Based on the historical violation rate, associated order density, amount deviation and time concentration of each graph node, the corresponding risk feature scalar is determined. The spatiotemporal correlation strength is weighted and aggregated by the risk feature scalar and edge relationship to obtain the local subgraph anomaly score centered on the current order. In actual evaluation, the invention utilizes the unique identification capabilities of the GS1 coding system (GTIN identifies the product category, AI 21 identifies the individual product serial number, GLN identifies the merchant / recycling point location, and AI 10 identifies the batch) to construct a spatiotemporal relationship diagram of the entire product lifecycle; The entities include product entities, merchant entities, consumer entities, order entities, old machine recycling entities, dismantling entities, and subsidy entities; in the node set, merchant nodes and dismantling nodes carry spatial coordinate attributes obtained by parsing GLN location encoding, and each node carries a business occurrence timestamp attribute; The edge set includes a sales edge, a logistics circulation edge, an old-for-new replacement edge, and a subsidy fund edge. In the edge set, the sales edge connects the order node with the product node and the merchant node, the logistics circulation edge connects the old machine recycling node with the dismantling node, the old-for-new replacement edge connects the consumer node with the old machine recycling node and the order node, and the subsidy fund edge connects the subsidy node with the order node and the consumer node. The spatiotemporal correlation strength is used to quantify the tightness of each correlation edge in the transaction chain in terms of time, space, and encoded correlation dimensions. The calculation formula is as follows: ; in For the edge The spatiotemporal correlation strength, edge Connecting nodes and , The time decay coefficient, For nodes and The time difference of business occurrence The spatial attenuation coefficient, For nodes and Euclidean spatial distance, For nodes and GS1 encoding between, Encode the sharing level indicator function for GS1; When two graph nodes simultaneously satisfy multiple GS1 code sharing conditions, the highest level in the indicator function is taken as the GS1 code sharing level of that edge; The local subgraph anomaly score is used to assess the overall anomaly level of the transaction network in which the current order is located, and the calculation formula is: ; in For orders Local subgraph centered Abnormal scores, The number of nodes in the subgraph. For nodes The k-th order neighborhood, For nodes Neighboring nodes, For nodes and Edge type weight, For the risk characteristics scalar of neighboring nodes, For nodes Node degree in a local subgraph; The edge type weight is preset according to the risk level of the business association type. Among them, the edge weight of the same AI 21 single product serial number associated with multiple subsidy nodes is the highest, the edge weight of the same GLN merchant associated with a large number of different consumers in a short period of time is the second highest, and the edge weight of different old machines flowing to the same dismantling enterprise through the same logistics trajectory is the third highest. In the risk characteristic scalar, the historical violation rate is the proportion of merchants or consumers belonging to the node that have been marked as abnormal in historical transactions; the associated order density is the number of orders associated with the node within a unit time window; the amount deviation is the degree of deviation between the amount of the associated transactions of the node and the filing benchmark price; and the time concentration is the degree of aggregation of the associated transactions of the node in a specific period.

[0021] In this embodiment, the method for obtaining the verification depth level and verification sampling frequency of the current order includes: Extract the real-time load of the verification system and merchant information, and calculate the real-time load coefficient of the system and the merchant risk profile score respectively; the real-time load coefficient of the system is the ratio of the current processing queue length to the maximum throughput of the system; the merchant risk profile score is calculated based on the historical compliance rate, complaint rate and return rate. Input the system's real-time load coefficient, merchant risk profile score, and local subgraph anomaly score into the verification depth decision function to obtain the verification depth level and verification sampling frequency for the current order. The expression is: ; ; in For orders The level of verification depth, For orders Local subgraph centered Abnormal scores, This is the threshold for the anomaly scoring benchmark. For merchants Risk profile scoring for Real-time load factor of the system For load sensitivity coefficient, This is a truncation function that cuts the output value to the interval between 1 and 3. For orders The verification sampling frequency, The maximum sampling frequency supported by the system. The sampling sensitivity coefficient, This is the threshold for abnormal scoring risk. It is a sigmoid activation function; In actual assessments, the verification depth level is used to classify orders into three different verification depths based on the degree of anomaly in the transaction chain and the system load status. In the verification depth decision function, when the real-time load coefficient of the system exceeds the preset threshold and the load sensitivity coefficient is greater than zero, the verification depth decision function reduces the verification depth level as the load increases, realizing automatic degradation processing during peak periods. When the real-time load coefficient of the system is lower than the preset low load threshold, the verification depth decision function increases the verification depth level as the load decreases, realizing deep verification processing during off-peak periods. In the verification depth decision function, the merchant risk profile score is obtained by weighting and aggregating the complaint rate and return rate based on the historical compliance rate. The expression is as follows: ; in Historical compliance rate As a weighting factor for complaints, To improve the complaint rate, For return weighting, Return rate; The verification sampling frequency is used to dynamically adjust the density of verification resources based on the transaction chain anomaly score; the higher the anomaly score, the higher the sampling frequency.

[0022] In this embodiment, the single-chain lightweight verification process only performs basic consistency comparison processing on the current transaction order in the areas of product identity verification, transaction certificate verification, consumer information verification, old device information verification, and merchant registration verification. In actual evaluation, the basic consistency comparison process involves performing a fast equivalence comparison or threshold judgment only on the core key fields of the current transaction order; the core key fields include the product identifier field, the transaction amount field, and the consumer identity identifier field; When verifying the identity of a product, it is necessary to verify the consistency of the entire chain between the new product identifier recorded by the sales platform, the old product recycling identifier recorded by the recycling platform, and the dismantling confirmation identifier recorded by the dismantling company. When verifying transaction vouchers, the verification is conducted independently from three sub-entities: the sales side, the recovery side, and the subsidy side, and then cross-verified. Specifically, the sales side verifies the consistency between sales vouchers and payment records, the recovery side verifies the consistency between recovery valuation and recovery payment, and the subsidy side verifies the consistency between subsidy eligibility records, write-off records, and fund disbursement records. When verifying consumer information, the consistency of the identity of the same consumer in the sales platform and the subsidy platform shall be verified. When verifying the information of used machines, the data of dismantling enterprises is used as the anchor point to verify the consistency of the time sequence logic between the old machine recycling time, dismantling confirmation time and subsidy application time, as well as whether the same old machine identifier is associated with multiple subsidies. When verifying merchant registration, the consistency between the merchant registration baseline information and the actual sales information is verified.

[0023] In this embodiment, the related transaction chain verification simultaneously performs single-chain lightweight verification processing on the current order and neighboring orders that are related to the current order.

[0024] In this embodiment, the full-volume deep review performs full consistency verification on all types of current transaction orders, including product identity verification, transaction voucher verification, consumer information verification, old device information verification, and merchant registration verification. After reviewing historical orders of the same type, the results are output to the manual final review queue. The historical order backtracking and verification involves retrieving historical orders that have the same merchant identifier, the same consumer identifier, or the same used product category identifier as the current order, and identifying abnormal historical orders across time dimensions. The historical similar order retrospective review adopts a strategy based on marginal risk and return, and calculates the marginal risk and return to iteratively solve for the optimal retrospective parameters to determine the retrospective range; the termination condition of the iterative solution is that the cumulative review cost reaches the daily manual review budget limit or the marginal risk and return is lower than the preset termination threshold. In actual evaluation, the full consistency verification process involves performing multi-field cross-verification, time-series logic verification, and historical statistical deviation calculation on all fields of the current transaction order. These fields include core key fields and auxiliary related fields. When reviewing historical orders of the same type, the review cost during the review process needs to meet the daily manual review budget constraint. The formula for calculating the marginal risk return is as follows: ; in For marginal risk and return, To trace the number of historical orders, To trace the time span, This is the additional risk amount detected during retrospective analysis. To cover the review costs during retrospective analysis, , For the first The number of historical orders and the time span corresponding to the group of backtracking parameters; The scope of historical orders of the same type is dynamically determined according to the following priorities: the first priority is recent orders from the same merchant and the same category of used equipment; the second priority is recent orders from the same consumer; and the third priority is orders with the same used equipment identifier. Candidate orders are accumulated in the order of the above priorities, and the backtracking stops when the cumulative review cost reaches the daily manual review budget or the marginal risk-reward is maximized.

[0025] In this embodiment, the method for optimizing the verification depth decision function based on the verification effect includes: Based on the verification results, a strategy effectiveness evaluation function is constructed. With the goal of maximizing the strategy effectiveness evaluation function, the dynamic parameters in the verification depth decision function are optimized online using a zero-order optimization method. The dynamic parameters include the anomaly scoring baseline threshold, the load sensitivity coefficient, the sampling sensitivity coefficient, and the anomaly scoring risk threshold. In actual evaluation, the expression for the strategy effectiveness evaluation function is as follows: ; in for Verification results within the time period This refers to the abnormal order interception rate. For false interception rate, To average verification costs, , , The benefit weight is used; the average verification cost is the normalized dimensionless relative cost in the calculation.

[0026] In addition, this invention also provides a set of implementation examples for verifying the consistency of trade-in orders for 55-inch televisions. Taking a trade-in order for a certain brand of 55-inch television as an example, the data is extracted from the Five Sources system through the standard API as follows: The sales platform extracts: product identification information (GTIN code, SN code), transaction voucher information (sales invoice image, POS payment record), and consumer information (ID card hash value, mobile phone number hash value). The recycling platform extracts: old device information (old device SN code, recycling valuation form, recycling payment voucher, logistics tracking number) and product identification information (old device GTIN code); The dismantling company extracts: information about the used machine (dismantling task confirmation form, dismantling process video, weight of dismantled products), and product identification information (used machine serial number). The subsidy platform extracts: transaction voucher information (unique identifier for subsidy eligibility, verification status code, fund disbursement record) and consumer information (identity identifier linked to the subsidy). Merchant registration system extracts: Merchant information (GLN code, benchmark price, energy efficiency registration level, business license qualification number, and historical violation record markers); A GS1 five-source semantic alignment tree is constructed, with the GS1 application identifier as the root node, extending downwards to data domain nodes (AI21 serial number field, AI 01 GTIN field), format nodes (such as pure numeric, alphanumeric mixed), and value domain constraint nodes (such as length constraints, check digit rules); leaf nodes are the specific field mapping targets under each data domain (such as IMEI sub-node, SN sub-node); the mapping form is to map the heterogeneous field names of each source system to the leaf nodes of the GS1 semantic alignment tree to achieve cross-system semantic equivalence; Take the mapping weights as 0.3 / 0.4 / 0.3, the high confidence threshold as 0.85, and the low confidence threshold as 0.40, and calculate the mapping confidence of field 1, field 2, and field 3 to the corresponding leaf nodes respectively; For the target leaf node of the sales platform field 1 and the AI ​​21 serial number field, the similarity of the field label characters is 0.9, the similarity of the value range distribution is 0.95 (the value range is 15 digits and meets the SN verification rules), the contextual conditional probability is 0.88, and the calculated mapping confidence is 0.914. Therefore, the sales platform field 1 is directly mapped to the target leaf node of the AI ​​21 serial number field. For the target leaf nodes of field 2 in the recycling platform and the quantity domain of AI 30, the similarity of field label characters is 0.2, the similarity of value range distribution is 0.3, the contextual conditional probability is 0.4, the mapping confidence is calculated to be 0.3, so the mapping of this field is rejected and it is marked as an unknown field; For the target leaf node n1 of field 3 of the recycling platform and the AI ​​21 serial number field, the similarity of field label characters is 0.5, the similarity of value range distribution is 0.92, the context conditional probability is 0.75, and the mapping confidence is calculated to be 0.743. The corresponding leaf node is selected as a candidate leaf node, and another candidate leaf node n2 of field 3 of the recycling platform is selected and entered into the disambiguation process. With a disambiguation coefficient of 0.1 / 0.2, taking the disambiguation score calculation of candidate leaf node n1 of field 3 of the recycling platform as an example, based on the depth of target leaf node n1 in the semantic tree of 3 and the number of competing nodes at the same level of 2, the disambiguation score is calculated to be 0.69. Similarly, the disambiguation score of candidate leaf node n2 is calculated to be 0.65. Select target leaf node n1 of AI 21 serial number field as the final mapping target of field 3 of the recycling platform. After completing the mapping of all fields according to the above method, the five-source standardized record set is obtained. Selecting a set of key semantic nodes This includes GTIN nodes, AI 21 serial number nodes, and price nodes, with corresponding business weights of 0.2 / 0.5 / 0.3 respectively (AI 21 serial number has the highest weight because it is directly associated with the risk of selling one device to multiple customers). The historical error penalty coefficients for each data source are set as follows: sales platform 0.05 (structured system, high reliability), recycling platform 0.15 (semi-manual data entry), dismantling enterprise 0.02 (automatic data collection from IoT devices, highest reliability), subsidy platform 0.08 (government system), and merchant registration system 0.20 (manual application). After cross-source vector alignment and credibility correction, the consistency kernel function outputs for each node are: GTIN node 0.95, AI21 node 0.88, and price node 0.92. The multi-source consistency metric is calculated to be 0.906 (the preset threshold is 0.8). The consistency of the five-source data is deemed acceptable, no cross-source data inconsistency alarm is triggered, and the current order verification process is terminated, proceeding to the next step of verification. A multi-source spatiotemporal correlation graph is constructed using order data from the sales platform as the link. The spatiotemporal correlation strength is calculated using the sales edge (edge ​​1) between order nodes and product nodes, and the logistics flow edge (edge ​​2) between old machine recycling nodes and dismantling nodes as examples (the time decay coefficient and spatial decay coefficient are taken as 0.1 and 0.01 respectively): Edge 1: The order node and the product node belong to the same order and occur within the same merchant. The time difference and spatial distance between the nodes are set to 0. The GS1 code sharing level is determined to be 0.5 based on the shared GTIN product category code. The spatiotemporal association strength of edge 1 is calculated to be 0.5. Edge 2: The old machine recycling time is 10:00 on August 1, 2026, and the dismantling confirmation time is 14:30 on August 3, 2026. The time difference is 2.2 days. The spatial distance based on GLN resolution is 50km. The SN code GS1 encoding sharing level of the same old machine is 1. The spatiotemporal correlation strength of edge 2 is calculated to be 0.487. Similarly, the spatiotemporal correlation strength of the remaining edges is calculated. The edge type weights are preset to 0.6 / 0.8 / 0.1 for the sales edge / logistics circulation edge / subsidy fund edge, and a single value (scalar) is taken for the risk feature scalar. The risk feature values ​​of the neighboring nodes of the merchant node / consumer node / dismantling node / subsidy node / old machine recycling node are 0.3 / 0.5 / 0.2 / 0.8 / 0.1, respectively. The spatiotemporal correlation strength, edge type weight and risk feature value of each edge are substituted into the formula and aggregated to obtain a local subgraph anomaly score of 0.736. Extract the real-time load of the verification system (current pending verification queue length is 300, system maximum throughput is 1000) and merchant information (historical compliance rate 0.90, complaint rate 0.05, return rate 0.03), take a complaint weight of 2 and a return weight of 1, calculate the system real-time load coefficient of 0.3 and the merchant risk profile score of 1, and input them into the verification depth decision function (abnormal score benchmark threshold of 0.55, load sensitivity coefficient of 0.5), output verification depth level of 3, trigger full-scale deep review; take the maximum sampling frequency of 100 (dimensionless calculation, actual dimension is times / second), sampling sensitivity coefficient of 2.0, abnormal score risk threshold of 0.50, calculate the verification sampling frequency as 61.6 times / second; Original order full consistency verification processing: Rapid equivalence comparison (the product SN code "SN20260801TV001" is completely consistent across the sales platform, recycling platform, and dismantling company records; the transaction amount of 2999 yuan is consistent across all three sources; the consumer's ID card hash value Hash_A1B2C3 is consistent), cross-verification (sales invoice amount 2999 yuan = POS payment record 2999 yuan = subsidy platform record amount 2999 yuan; old machine recycling valuation 200 yuan = recycling payment voucher 200 yuan; subsidy amount 600 yuan = fund disbursement record 600 yuan), time-sequence logic verification (old machine recycling time (2026-08-01 10:00) < dismantling confirmation time (2026-08-03 14:30) < subsidy application time (2026-08-03) 16:00), satisfying the time sequence constraint), historical statistical deviation (current selling price is 2999 yuan, merchant filing benchmark price is 3199 yuan, deviation = (3199-2999) / 3199 = 6.25%, which is lower than the 10% preset threshold); The daily manual review budget is set to a maximum of 20 orders, and the scope of backtracking is dynamically determined according to priority: 3 historical orders are extracted for the first priority (same merchant and same TV product category, within the last 7 days); 2 historical orders are extracted for the second priority (same consumer, within the last 30 days); and 1 historical order is extracted for the third priority (same old machine SN code, within the last 90 days). A total of 6 historical orders were extracted. The total review cost did not exceed the budget. When continuing to trace back, the marginal risk-reward decreased to 0.02, which is lower than the preset termination threshold of 0.05. Therefore, the backtracking was stopped. After performing a full consistency check on the above 6 historical orders, they were output to the manual final review queue for manual review. After extracting the cumulative verification data (100 verified orders) after 1 hour: abnormal order interception rate 0.15, false interception rate 0.02, average verification cost 0.85. Taking the benefit weights as 3 / 5 / 1, the calculated strategy effect evaluation function is -0.5, indicating that the strategy effect is negative, which means that the current parameter settings are too conservative. The dynamic parameters in the verification depth decision function were optimized online using a zero-order optimization method: the anomaly scoring baseline threshold was adjusted from 0.55 to 0.50 (to improve anomaly sensitivity), the load sensitivity coefficient was adjusted from 0.5 to 0.6 (to enhance load response), the sampling sensitivity coefficient was adjusted from 2.0 to 2.2 (to enhance sampling response), and the anomaly scoring risk threshold was adjusted from 0.50 to 0.45 (to reduce the risk threshold). The optimized verification depth decision function achieves more accurate risk-resource matching in subsequent order processing.

[0027] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for verifying the consistency of trade-in product data based on the GS1 coding rule engine, characterized in that, Includes the following steps: S1. Access the multi-source heterogeneous data of the old-for-new business through the standard API, and perform semantic mapping and standardization processing on the fields of each source data through the GS1 semantic alignment tree to obtain a multi-source standardized record set. S2. Calculate the multi-source consistency metric based on the multi-source standardized record set and perform cross-source data consistency verification; S3. Construct a multi-source spatiotemporal association graph based on the GS1 encoding in the multi-source standardized record set as the association link, and calculate the spatiotemporal association strength of each edge in the graph and the anomaly score of the local subgraph centered on the current order. S4. Extract the real-time load and merchant information of the verification system, and combine the local subgraph anomaly score with the input verification depth decision function to obtain the verification depth level and verification sampling frequency of the current order; S5. Perform differentiated verification based on the current order's verification depth level and output the verification conclusion. Optimize the verification depth decision function based on the verification results. The multi-source heterogeneous data includes product identification information, transaction voucher information, consumer information, used device information, and merchant information; the multi-source heterogeneous data comes from sales platforms, recycling platforms, dismantling companies, subsidy platforms, and merchant registration systems; The differentiated verification includes single-chain lightweight verification processing, related transaction chain verification, and full-scale in-depth review.

2. The data consistency verification method for trade-in products based on the GS1 coding rule engine according to claim 1, characterized in that, The method for obtaining a multi-source normalized recordset includes: Construct a GS1 five-source semantic alignment tree, with the GS1 application identifier as the root node, and expand downwards to data domain nodes, format nodes, and value domain constraint nodes. Calculate the mapping confidence of each source data field to all leaf nodes of the GS1 semantic tree. When the mapping confidence is greater than the high confidence threshold, the corresponding source data field is directly mapped to the corresponding leaf node. When the mapping confidence is less than the low confidence threshold, the mapping of the corresponding source data field is rejected. Otherwise, the corresponding leaf node is used as a candidate leaf node for disambiguation processing. The disambiguation process calculates the disambiguation score of each candidate leaf node using a hierarchical disambiguation function, and selects the node with the highest disambiguation score as the final mapping target. Complete the semantic mapping of all source data fields to obtain a multi-source standardized record set; The formula for calculating the mapping confidence level is: ； in For fields To the leaf node The mapping confidence, , , For mapping weights, For data source Middle field The tag, Leaf nodes in the GS1 semantic tree The tag, For the similarity of field label characters, For fields Value range distribution, For nodes The standard range For the similarity of value range distribution, For fields Contextual information, The semantic node conditional probability under context conditions; The disambiguation score is calculated as follows: ； in For nodes The disambiguation score, , The disambiguation coefficient is... For nodes Depth in the semantic tree For nodes The number of competing nodes at the same level.

3. The data consistency verification method for trade-in products based on the GS1 coding rule engine according to claim 1, characterized in that, The method for cross-source data consistency verification includes: A multi-source consistency metric is calculated based on a multi-source standardized record set. Cross-source data consistency is then checked based on the multi-source consistency metric. If the multi-source consistency metric is lower than a preset threshold, a cross-source data inconsistency alarm is triggered and the current order verification process is terminated. Otherwise, the next step of verification is performed. The multi-source consistency metric is used to evaluate the degree of consistency of multi-source data at key semantic nodes. It is determined by calculating the node consistency of each data source at the corresponding node using the Gaussian radial basis kernel function and weighting the results.

4. The method for verifying the consistency of trade-in product data based on the GS1 encoding rule engine according to claim 1 or 3, characterized in that, The method for calculating the spatiotemporal correlation strength of each edge in the graph and the anomaly score of the local subgraph centered on the current order includes: The entities in the multi-source standardized record set are abstracted into graph nodes to form a node set. The graph edges are constructed based on the business relationships of the old-for-new business to form an edge set. The spatiotemporal correlation strength of each edge in the graph is calculated based on the time difference, spatial distance, and GS1 coding sharing level between each graph node. The spatiotemporal correlation strength is negatively correlated with the time difference and spatial distance, and positively correlated with the GS1 coding sharing level. The GS1 coding sharing level is determined by the GS1 coding sharing level indicator function. The GS1 code sharing level indicator function is characterized by the following: it takes 1 when two graph nodes share the same AI 21 single item serial number, 0.8 when two graph nodes share the same AI 10 production batch number, 0.5 when two graph nodes share the same GTIN product category code but have different single item serial numbers, and 0 when two graph nodes have no GS1 code association. Based on the historical violation rate, related order density, amount deviation and time concentration of each graph node, the corresponding risk feature scalar is determined. The spatiotemporal correlation strength is weighted and aggregated by the risk feature scalar and edge relationship to obtain the local subgraph anomaly score centered on the current order.

5. The method for verifying the consistency of trade-in product data based on the GS1 coding rule engine according to claim 1, characterized in that, The method for obtaining the verification depth level and verification sampling frequency of the current order includes: Extract the real-time load of the verification system and merchant information, and calculate the real-time load coefficient of the system and the merchant risk profile score respectively; the real-time load coefficient of the system is the ratio of the current processing queue length to the maximum throughput of the system; the merchant risk profile score is calculated based on the historical compliance rate, complaint rate and return rate. Input the system's real-time load coefficient, merchant risk profile score, and local subgraph anomaly score into the verification depth decision function to obtain the verification depth level and verification sampling frequency for the current order. The expression is: ； ； in For orders The level of verification depth, For orders Local subgraph centered Abnormal scores, This is the threshold for the anomaly scoring benchmark. For merchants Risk profile scoring for Real-time load factor of the system For load sensitivity coefficient, This is a truncation function that cuts the output value to the interval between 1 and 3. For orders The verification sampling frequency, The maximum sampling frequency supported by the system. The sampling sensitivity coefficient, This is the threshold for abnormal scoring risk. It is a sigmoid activation function.

6. The method for verifying the consistency of trade-in product data based on the GS1 coding rule engine according to claim 1, characterized in that: The single-chain lightweight verification process only performs basic consistency comparison processing on the current transaction order in the areas of product identity verification, transaction certificate verification, consumer information verification, old machine information verification, and merchant registration verification.

7. The method for verifying the consistency of trade-in product data based on the GS1 coding rule engine according to claim 6, characterized in that: The related transaction chain verification also performs lightweight single-chain verification processing on the current order and neighboring orders that are related to the current order.

8. The method for verifying the consistency of trade-in product data based on the GS1 coding rule engine according to claim 6, characterized in that: The full-volume in-depth review performs full consistency verification on all types of current transaction orders, including product identity verification, transaction voucher verification, consumer information verification, old device information verification, and merchant registration verification. After reviewing historical orders of the same type, the results are output to the manual final review queue. The historical order backtracking and verification involves retrieving historical orders that have the same merchant identifier, the same consumer identifier, or the same used product category identifier as the current order, and identifying abnormal historical orders across time dimensions. The historical similar order retrospective review adopts a strategy based on marginal risk and return, and calculates the marginal risk and return to iterate and solve for the optimal retrospective parameters to determine the retrospective range; the termination condition of the iterative solution is that the cumulative review cost reaches the daily manual review budget limit or the marginal risk and return is lower than the preset termination threshold.

9. The method for verifying the consistency of trade-in product data based on the GS1 coding rule engine according to claim 1, characterized in that, The method for optimizing the verification depth decision function based on the verification effect includes: Based on the verification results, a strategy effectiveness evaluation function is constructed. With the goal of maximizing the strategy effectiveness evaluation function, the dynamic parameters in the verification depth decision function are optimized online using a zero-order optimization method. The dynamic parameters include the anomaly scoring benchmark threshold, the load sensitivity coefficient, the sampling sensitivity coefficient, and the anomaly scoring risk threshold.

Images