Cross-border e-commerce risk identification method and related product
By combining multi-dimensional feature extraction and model recognition with manual review, the accuracy problem of risk identification in cross-border e-commerce has been solved, achieving efficient risk point identification and improved accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-03-24
AI Technical Summary
Traditional cross-border e-commerce risk identification methods rely on manual review, resulting in incomplete risk identification coverage, delayed response, and difficulty in obtaining real operating data from overseas platforms, thus affecting accuracy.
By acquiring the operational data of the stores to be identified, multi-dimensional feature extraction is performed, including store sales, expenses, information and compliance risk features. The risk identification is carried out using a gradient boosting decision tree classification model and a scorecard model, and secondary verification is performed by combining manual review and a rule engine.
It improves the accuracy of risk identification in cross-border e-commerce, avoids misjudgments and omissions, covers the entire business picture, and achieves efficient risk point identification.
Smart Images
Figure CN121724629A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a method for identifying risks in cross-border e-commerce and related products. Background Technology
[0002] With the rapid development of global e-commerce, cross-border e-commerce has become an important bridge connecting domestic and international markets. More and more SMEs are conducting cross-border trade through overseas e-commerce platforms, driving the digital transformation of international trade. However, while bringing vast market opportunities, cross-border e-commerce also carries significant operational risks.
[0003] Due to the characteristics of cross-border e-commerce, such as high transaction frequency, complex supply chain links, and long fund settlement cycle, traditional cross-border e-commerce risk identification methods mainly rely on manual methods. However, there are obvious shortcomings in manually reviewing business information such as store registration information and bank statements: the operating data of cross-border e-commerce, such as orders, logistics, inventory, and user reviews, are stored on overseas platforms, making it difficult for domestic institutions to obtain real and complete business information. This results in incomplete risk identification coverage and delayed response, which seriously affects the accuracy of cross-border e-commerce risk identification.
[0004] Improving the accuracy of risk identification in cross-border e-commerce is a technical problem that urgently needs to be solved. Summary of the Invention
[0005] To address the aforementioned issues, this application provides a method and related products for identifying risks in cross-border e-commerce, with the aim of improving the accuracy of risk identification in cross-border e-commerce.
[0006] The embodiments of this application disclose the following technical solutions:
[0007] The first aspect of this application provides a method for identifying risks in cross-border e-commerce, the method comprising:
[0008] Obtain the operational data of the store to be identified;
[0009] Multi-dimensional feature extraction is performed on the operating data of the store to be identified to obtain multi-dimensional feature extraction results; the multi-dimensional feature extraction includes store sales feature extraction, expense feature extraction, store information feature extraction, and store compliance risk feature extraction.
[0010] Based on the multi-dimensional feature extraction results, the risks of the stores to be identified are determined, and the risk identification results are obtained.
[0011] Optionally, the risk identification of the store to be identified based on the multi-dimensional feature extraction results, to obtain risk identification results, includes:
[0012] Based on the multi-dimensional feature extraction results, the risk prediction probability value of the store to be identified is determined;
[0013] The risk prediction probability value of the store to be identified is converted to obtain the risk score of the store to be identified.
[0014] The risk level of the store to be identified is determined based on its risk score.
[0015] Optionally, determining the risk prediction probability value of the store to be identified based on the multi-dimensional feature extraction results includes:
[0016] Based on the multi-dimensional feature extraction results, multiple gradient boosting decision tree classification models are used for prediction to obtain multiple risk prediction results; the multiple gradient boosting decision tree classification models are obtained after training.
[0017] The multiple risk prediction results are processed to obtain the risk prediction probability value of the store to be identified.
[0018] Optionally, the training methods for the multiple gradient boosting decision tree classification models include:
[0019] The importance of features in multiple candidate dimensions is quantified based on historical sample data to obtain the quantification result of features in each candidate dimension; the historical sample data includes positive sample data and negative sample data; the positive sample data and the negative sample data are obtained by processing the historical sample data using a clustering algorithm;
[0020] Based on the quantization results of features in multiple candidate dimensions and a preset threshold, features in multiple candidate dimensions are filtered to obtain features in multiple target dimensions.
[0021] Multiple sets of hyperparameter configurations are determined using Bayesian parameter tuning.
[0022] Multiple gradient boosting decision tree classification models are obtained by training based on the features of the multiple target dimensions and the multiple sets of hyperparameter configurations.
[0023] Optionally, after determining the risk level of the store to be identified based on its risk score, the method further includes:
[0024] If the risk level of the store to be identified meets the preset conditions, the store to be identified will be manually reviewed; if the risk level of the store to be identified does not meet the preset conditions, the store to be identified will be reviewed using a preset rule engine.
[0025] Optionally, the conversion of the risk prediction probability value of the store to be identified to obtain the risk score of the store to be identified includes:
[0026] The risk prediction probability value of the store to be identified is transformed using a scorecard model to obtain the risk score of the store to be identified; the scorecard model is used to perform nonlinear transformation on the probability value based on preset scorecard parameters to obtain the score value within a preset range.
[0027] A second aspect of this application provides a cross-border e-commerce risk identification device, the device comprising:
[0028] The data acquisition module is used to acquire the operating data of the stores to be identified;
[0029] The feature extraction module is used to extract multi-dimensional features from the operating data of the store to be identified, and obtain multi-dimensional feature extraction results.
[0030] The risk identification module is used to identify the risks of the store to be identified based on the multi-dimensional feature extraction results, and obtain the risk identification results.
[0031] Optionally, the risk identification module is used for:
[0032] Based on the multi-dimensional feature extraction results, the risk prediction probability value of the store to be identified is determined;
[0033] The risk prediction probability value of the store to be identified is converted to obtain the risk score of the store to be identified.
[0034] The risk level of the store to be identified is determined based on its risk score.
[0035] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the cross-border e-commerce risk identification method provided in any implementation of the first aspect.
[0036] The fourth aspect of this application provides a processor for running a computer program that, when running, executes the cross-border e-commerce risk identification method provided in any implementation of the first aspect.
[0037] Compared with the prior art, this application has the following beneficial effects:
[0038] The cross-border e-commerce risk identification method provided in this application embodiment obtains the operating data of the store to be identified. Multi-dimensional feature extraction is performed on the operating data of the store to be identified to obtain multi-dimensional feature extraction results. The multi-dimensional feature extraction includes feature extraction of store sales, expense information, store information, and compliance risk. Based on the multi-dimensional feature extraction results, the risks of the store to be identified are identified to obtain risk identification results. By performing multi-dimensional feature extraction on the operating data of the store to be identified and identifying risks based on the multi-dimensional feature extraction results, misjudgments and omissions caused by relying on a single indicator are avoided. This method covers the entire business picture, effectively identifies risk points, and thus improves the accuracy of cross-border e-commerce risk identification. Attached Figure Description
[0039] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0040] Figure 1 A flowchart illustrating a cross-border e-commerce risk identification method provided in this application embodiment;
[0041] Figure 2 This is a schematic diagram of the structure of a cross-border e-commerce risk identification device provided in an embodiment of this application. Detailed Implementation
[0042] As described earlier, due to the characteristics of cross-border e-commerce, such as high transaction frequency, numerous and complex supply chain links, and long fund settlement cycles, traditional cross-border e-commerce risk identification methods mainly rely on manual methods. However, manual review of merchant registration information and bank statements has obvious shortcomings: on the one hand, manual review is inefficient and difficult to handle complex cross-border transactions; on the other hand, because the operational data of cross-border e-commerce, such as orders, logistics, inventory, and user reviews, are stored on overseas platforms, domestic institutions have difficulty obtaining real and complete business information, resulting in incomplete risk identification coverage and delayed response, which seriously affects the accuracy of cross-border e-commerce risk identification.
[0043] In view of the above problems, this application proposes a method and related products for identifying risks in cross-border e-commerce, which involves obtaining the operating data of the store to be identified; performing multi-dimensional feature extraction on the operating data of the store to be identified to obtain multi-dimensional feature extraction results; the multi-dimensional feature extraction includes store sales feature extraction, expense feature extraction, store information feature extraction, and store compliance risk feature extraction; and identifying the risks of the store to be identified based on the multi-dimensional feature extraction results to obtain risk identification results.
[0044] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.
[0045] See Figure 1 This figure is a flowchart of a cross-border e-commerce risk identification method provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps:
[0046] S101. Obtain the operating data of the store to be identified.
[0047] This involves data authorization through cross-border e-commerce, directly connecting to the data interface of the cross-border e-commerce platform via technical integration to obtain the store's operational data.
[0048] By directly connecting to cross-border e-commerce platforms to obtain store operating data, bypassing manual entry or third-party transcription, problems such as data tampering or providing false data are avoided, ensuring the authenticity and authority of the data. This provides a reliable data foundation for subsequent multi-dimensional feature extraction and risk identification, and helps improve the accuracy of cross-border e-commerce risk identification.
[0049] S102. Perform multi-dimensional feature extraction on the operating data of the store to be identified to obtain multi-dimensional feature extraction results.
[0050] The multi-dimensional feature extraction includes features extraction of store sales, features extraction of expenses, features extraction of store information, and features extraction of store compliance risks.
[0051] Store sales data includes monthly sales revenue, order volume, average order value, return rate, or repurchase rate; expense data includes the percentage of marketing expenses, fluctuations in logistics costs, or platform commissions; store information includes registration duration, frequency of legal representative changes, or number of associated stores; and store compliance risks include the number of infringement complaints, false shipping records, or missing qualifications.
[0052] By extracting features from multiple dimensions, a cross-validation mechanism can be formed, reducing false positive and false negative rates, providing a high-quality data foundation for subsequent risk identification, and helping to improve the accuracy of risk identification in cross-border e-commerce.
[0053] S103. Based on the multi-dimensional feature extraction results, the risks of the store to be identified are identified to obtain risk identification results.
[0054] In one feasible implementation:
[0055] Based on the multi-dimensional feature extraction results, the risk prediction probability value of the store to be identified is determined;
[0056] The risk prediction probability value of the store to be identified is converted to obtain the risk score of the store to be identified.
[0057] The risk level of the store to be identified is determined based on its risk score.
[0058] In one optional implementation, based on the multi-dimensional feature extraction results, determining the risk prediction probability value of the store to be identified includes:
[0059] Based on the multi-dimensional feature extraction results, multiple gradient boosting decision tree classification models are used for prediction to obtain multiple risk prediction results; the multiple gradient boosting decision tree classification models are obtained after training.
[0060] The multiple risk prediction results are processed to obtain the risk prediction probability value of the store to be identified.
[0061] By using multiple trained gradient boosting decision tree classification models to make parallel predictions for the same store, and then fusing the multiple prediction results, a unified risk prediction probability value is finally output. Multiple models learn data patterns from different perspectives, avoiding misjudgment caused by a single model having too high confidence, making the final risk prediction probability value closer to the real risk distribution, which helps to improve the accuracy of risk prediction for the store to be identified.
[0062] In one optional implementation, the training method of the plurality of gradient boosting decision tree classification models includes:
[0063] The importance of features in multiple candidate dimensions is quantified based on historical sample data to obtain the quantification result of features in each candidate dimension; the historical sample data includes positive sample data and negative sample data; the positive sample data and the negative sample data are obtained by processing the historical sample data using a clustering algorithm.
[0064] Based on the quantization results of features in multiple candidate dimensions and a preset threshold, features in multiple candidate dimensions are filtered to obtain features in multiple target dimensions.
[0065] Bayesian parameter tuning is used to determine multiple sets of hyperparameter configurations.
[0066] Multiple gradient boosting decision tree classification models are obtained by training based on the features of the multiple target dimensions and the multiple sets of hyperparameter configurations.
[0067] There are no restrictions on the clustering algorithm here. Taking the K-means clustering algorithm as an example: Based on historical business data, the K-means clustering algorithm is used to cluster positive and negative samples. The samples are clustered into K classes by setting the K value to obtain the clustering results. According to the clustering results, in the classes where labeled positive samples are clustered, the unlabeled samples in the classes are also mapped as positive samples, thereby increasing the proportion of positive samples.
[0068] Since the proportion of positive samples in actual business is very low and the ratio of positive to negative samples is unbalanced, a clustering algorithm is adopted here to cluster the samples. Samples that are clustered in the same class as the already labeled positive samples are also labeled as positive samples for model training, thereby reducing the problem of imbalance between positive and negative samples.
[0069] The store operation data undergoes data cleaning, standardization, and normalization, followed by data integration to obtain the data fields for model design. Based on business understanding and data distribution, basic and derived fields related to store operations are statistically analyzed for different time windows, such as 3 months, 6 months, 12 months, or 24 months. For dimensions such as store sales, expenses, store information, and store compliance risks, indicators within each dimension are calculated, along with metrics such as information value or Pearson correlation coefficients, to select features and ensure the stability and effectiveness of the features input to the model.
[0070] In one optional implementation, after determining the risk level of the store to be identified based on its risk score, the method further includes:
[0071] If the risk level of the store to be identified meets the preset conditions, the store to be identified will be manually reviewed; if the risk level of the store to be identified does not meet the preset conditions, the store to be identified will be reviewed using a preset rule engine.
[0072] The default condition is that the risk level of the store to be identified is higher than the preset risk level. The risk level of the store to be identified is determined by setting a risk score threshold and the correspondence between the risk score threshold and the preset risk level.
[0073] If the risk level of the store to be identified does not meet the preset conditions, the store to be identified will be reviewed using a preset rule engine. Expert experience rules will be used to make a second judgment on whether there is any relevant risk. If there is no risk, the process will proceed to the automatic approval process. If there is a certain risk, the process will return to the manual review process.
[0074] In one optional implementation, the conversion of the risk prediction probability value of the store to be identified to obtain the risk score of the store to be identified includes:
[0075] The risk prediction probability value of the store to be identified is transformed using a scorecard model to obtain the risk score of the store to be identified; the scorecard model is used to perform nonlinear transformation on the probability value based on preset scorecard parameters to obtain the score value within a preset range.
[0076] The formula for transformation using the scorecard model is as follows:
[0077] ;
[0078] ;
[0079] in, This represents the risk score, with a range of 300-900. This indicates the probability of the target event occurring; PDO indicates that when the odds double, the risk score will decrease by PDO points. Indicates the probability ratio The baseline score is set below. During application, PDO needs to be set manually. and Values, for example, pdo=50, , Different scoring card parameters can be set according to different actual scenarios, data distributions, and business requirements.
[0080] By using a scorecard model to convert abstract risk probability values into risk scores, the risk identification results are made readable. A standardized and unified risk measurement system is established, enabling multi-level risk response and supporting subsequent decision-making.
[0081] After obtaining the risk identification results of the stores to be identified, such as risk scores and risk levels, high-risk stores are alerted to remind business personnel to conduct further reviews, which can effectively reduce the risk of fraud.
[0082] The cross-border e-commerce risk identification method provided in this application embodiment obtains the operating data of the store to be identified. Multi-dimensional feature extraction is performed on the operating data of the store to be identified to obtain multi-dimensional feature extraction results. The multi-dimensional feature extraction includes feature extraction of store sales, expense information, store information, and compliance risk. Based on the multi-dimensional feature extraction results, the risks of the store to be identified are identified to obtain risk identification results. By performing multi-dimensional feature extraction on the operating data of the store to be identified and identifying risks based on the multi-dimensional feature extraction results, misjudgments and omissions caused by relying on a single indicator are avoided. This method covers the entire business picture, effectively identifies risk points, and thus improves the accuracy of cross-border e-commerce risk identification.
[0083] Based on the cross-border e-commerce risk identification method described in the preceding embodiments, this application also provides a cross-border e-commerce risk identification device. Figure 2 This is a schematic diagram of the device. Figure 2 As shown, the cross-border e-commerce risk identification device includes:
[0084] The data acquisition module 201 is used to acquire the operating data of the store to be identified;
[0085] The feature extraction module 202 is used to perform multi-dimensional feature extraction on the operating data of the store to be identified, and obtain multi-dimensional feature extraction results;
[0086] The risk identification module 203 is used to identify the risks of the store to be identified based on the multi-dimensional feature extraction results, and obtain the risk identification results.
[0087] As an optional embodiment, the risk identification module is used for:
[0088] Based on the multi-dimensional feature extraction results, the risk prediction probability value of the store to be identified is determined;
[0089] The risk prediction probability value of the store to be identified is converted to obtain the risk score of the store to be identified.
[0090] The risk level of the store to be identified is determined based on its risk score.
[0091] As an optional embodiment, determining the risk prediction probability value of the store to be identified based on the multi-dimensional feature extraction results includes:
[0092] Based on the multi-dimensional feature extraction results, multiple gradient boosting decision tree classification models are used for prediction to obtain multiple risk prediction results; the multiple gradient boosting decision tree classification models are obtained after training.
[0093] The multiple risk prediction results are processed to obtain the risk prediction probability value of the store to be identified.
[0094] As an optional embodiment, the training method of the plurality of gradient boosting decision tree classification models includes:
[0095] The importance of features in multiple candidate dimensions is quantified based on historical sample data to obtain the quantification result of features in each candidate dimension; the historical sample data includes positive sample data and negative sample data; the positive sample data and the negative sample data are obtained by processing the historical sample data using a clustering algorithm;
[0096] Based on the quantization results of features in multiple candidate dimensions and a preset threshold, features in multiple candidate dimensions are filtered to obtain features in multiple target dimensions.
[0097] Multiple sets of hyperparameter configurations are determined using Bayesian parameter tuning.
[0098] Multiple gradient boosting decision tree classification models are obtained by training based on the features of the multiple target dimensions and the multiple sets of hyperparameter configurations.
[0099] As an optional embodiment, after determining the risk level of the store to be identified based on its risk score, the method further includes:
[0100] If the risk level of the store to be identified meets the preset conditions, the store to be identified will be manually reviewed; if the risk level of the store to be identified does not meet the preset conditions, the store to be identified will be reviewed using a preset rule engine.
[0101] As an optional embodiment, the conversion of the risk prediction probability value of the store to be identified to obtain the risk score of the store to be identified includes:
[0102] The risk prediction probability value of the store to be identified is transformed using a scorecard model to obtain the risk score of the store to be identified; the scorecard model is used to perform nonlinear transformation on the probability value based on preset scorecard parameters to obtain the score value within a preset range.
[0103] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the cross-border e-commerce risk identification method as described in any of the method embodiments.
[0104] Furthermore, this application embodiment also provides a processor for running a computer program, which executes the cross-border e-commerce risk identification method as described in any of the foregoing method embodiments.
[0105] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. The components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment solution according to actual needs. Those skilled in the art can understand and implement this without creative effort.
[0106] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for identifying risks in cross-border e-commerce, characterized in that, include: Obtain the operational data of the store to be identified; Multi-dimensional feature extraction is performed on the operating data of the store to be identified to obtain multi-dimensional feature extraction results; the multi-dimensional feature extraction includes store sales feature extraction, expense feature extraction, store information feature extraction, and store compliance risk feature extraction. Based on the multi-dimensional feature extraction results, the risks of the stores to be identified are determined, and the risk identification results are obtained.
2. The method according to claim 1, characterized in that, The risk identification of the store to be identified based on the multi-dimensional feature extraction results yields the following results: Based on the multi-dimensional feature extraction results, the risk prediction probability value of the store to be identified is determined; The risk prediction probability value of the store to be identified is converted to obtain the risk score of the store to be identified. The risk level of the store to be identified is determined based on its risk score.
3. The method according to claim 2, characterized in that, The step of determining the risk prediction probability value of the store to be identified based on the multi-dimensional feature extraction results includes: Based on the multi-dimensional feature extraction results, multiple gradient boosting decision tree classification models are used for prediction to obtain multiple risk prediction results; the multiple gradient boosting decision tree classification models are obtained after training. The multiple risk prediction results are processed to obtain the risk prediction probability value of the store to be identified.
4. The method according to claim 3, characterized in that, The training methods for the multiple gradient boosting decision tree classification models include: The importance of features in multiple candidate dimensions is quantified based on historical sample data to obtain the quantification result of features in each candidate dimension; the historical sample data includes positive sample data and negative sample data; the positive sample data and the negative sample data are obtained by processing the historical sample data using a clustering algorithm; Based on the quantization results of features in multiple candidate dimensions and a preset threshold, features in multiple candidate dimensions are filtered to obtain features in multiple target dimensions. Multiple sets of hyperparameter configurations are determined using Bayesian parameter tuning. Multiple gradient boosting decision tree classification models are obtained by training based on the features of the multiple target dimensions and the multiple sets of hyperparameter configurations.
5. The method according to claim 2, characterized in that, After determining the risk level of the store to be identified based on its risk score, the process further includes: If the risk level of the store to be identified meets the preset conditions, the store to be identified will be manually reviewed; if the risk level of the store to be identified does not meet the preset conditions, the store to be identified will be reviewed using a preset rule engine.
6. The method according to claim 2, characterized in that, The process of converting the predicted risk probability value of the store to be identified into a risk score for the store to be identified includes: The risk prediction probability value of the store to be identified is transformed using a scorecard model to obtain the risk score of the store to be identified; the scorecard model is used to perform nonlinear transformation on the probability value based on preset scorecard parameters to obtain the score value within a preset range.
7. A cross-border e-commerce risk identification device, characterized in that, include: The data acquisition module is used to acquire the operating data of the stores to be identified; The feature extraction module is used to extract multi-dimensional features from the operating data of the store to be identified, and obtain multi-dimensional feature extraction results. The risk identification module is used to identify the risks of the store to be identified based on the multi-dimensional feature extraction results, and obtain the risk identification results.
8. The apparatus according to claim 7, characterized in that, The risk identification module is used for: Based on the multi-dimensional feature extraction results, the risk prediction probability value of the store to be identified is determined; The risk prediction probability value of the store to be identified is converted to obtain the risk score of the store to be identified. The risk level of the store to be identified is determined based on its risk score.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the cross-border e-commerce risk identification method as described in any one of claims 1-6.
10. A processor, characterized in that, Used to run a computer program, which, when running, executes the cross-border e-commerce risk identification method as described in any one of claims 1-6.