E-commerce transaction security assessment method and system based on big data

Abstract

The invention discloses an e-commerce transaction security assessment method and system based on big data, and relates to the technical field of transaction security, and the method comprises the following steps: S1, extracting commodity features, S2, extracting user features, S3, predicting a fraud probability, S4, determining a cost factor, and S5, taking related measures. According to the method, the fraud probability and the cost factor are combined during decision making, quantitative balance of the risk and the cost is achieved, decision making is intelligent and meets actual business appeals, rigidity caused by simply looking at the probability is avoided, meanwhile, the return behavior mode of the user is analyzed regularly, the long-term credibility of the user is evaluated dynamically, and the user experience is improved. And key links such as feature extraction, fraud prediction, cost factor calculation and the like are mutually independent, the high-cohesion and low-coupling structure significantly improves the overall robustness, when any single module needs to be optimized, the overall structure does not need to be reconstructed, and the maintenance cost is greatly reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of transaction security, in particular to an e-commerce transaction security evaluation method and system based on big data. BACKGROUND

[0002] The e-commerce transaction security evaluation method based on big data can accurately identify return fraud and quantify risk costs by constructing an intelligent multi-dimensional risk control system, providing an efficient and adaptive protection barrier for e-commerce. In the invention patent with the application number 202410910096.9, an online order processing method and system for an ERP system are disclosed, which includes the following steps: receiving a return request sent by a user, and sorting the display windows of the return requests sent by each user; according to the sorted display windows, the return requests are audited in turn to determine whether they meet the return standard; when the return request is approved, it is preliminarily determined whether the returned goods can be resold based on the return reason; when the returned goods can be resold, the matching degree of each warehouse and the returned goods is determined according to the warehouse information of the merchant and the information carried in the return request; the target warehouse and the target return address are determined according to the matching degree of each warehouse and the returned goods; when the target warehouse receives the returned goods and it is determined that the returned goods can be resold, the inventory data in the target warehouse is updated. The present application realizes the secondary optimization and utilization of returned goods, effectively alleviates the problem of overstock and shortage, and improves the business efficiency.

[0003] The above-mentioned prior art solves the problem of being unable to preliminarily determine whether the returned goods can be resold, but the system has low accuracy in identifying returned goods, resulting in high fraud loss and misjudgment rate. At the same time, the decision-making is rigid and cannot balance the risk and cost, causing business loss or customer loss. The system is highly dependent on manual auditing, which is inefficient and slow in response. Moreover, the modules are tightly coupled, the maintenance and upgrade cost is high, the user behavior cannot be dynamically evaluated, and it is difficult to deal with new fraud methods. SUMMARY

[0004] The present application aims to provide an e-commerce transaction security evaluation method and system based on big data to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solution: an e-commerce transaction security evaluation method based on big data, comprising the following steps:

[0006] S1, extracting the characteristics of the goods: reading the measured volume, standard volume, state score of the packaging components and color difference value of the specified returned goods, and analyzing the volume characteristics, weight characteristics, packaging characteristics and color characteristics of the goods;

[0007] S2. Extract user features: Read all return records of the user corresponding to the product, and calculate the fraud features, return frequency features and deadline approach features of the current user based on the records;

[0008] S3. Predict fraud probability: After statistically analyzing the capacity, weight, packaging, and color characteristics of the product, determine the fraud characteristics, return frequency characteristics, and deadline approach characteristics of the user corresponding to the product. After normalizing all characteristics, input them into the deployed LightGBM model and output the predicted fraud probability value corresponding to the current product.

[0009] S4. Determine the cost factor: Read the product category, user credibility, and current review queue length, and calculate the cost factor based on the product category, user credibility, current review queue length, and return records;

[0010] S5. Take relevant measures: Determine the zoning based on the predicted fraud probability value and cost factors of the goods, take corresponding measures according to the zoning, assign the return information to the auditors who are to be manually reviewed, and analyze the credibility of different users.

[0011] Preferably, step S1 specifically includes the following steps:

[0012] S101, Read the... individual products Actual capacity of returned goods and standard capacity Then, using and Calculate capacity characteristics ,in Read Actual weight of returned goods and standard weight Then, using and Calculate the standard weight characteristics ,in , Indicates the serial number;

[0013] S102, Read the... individual products Weighting coefficients for all packaging components and status score ,according to and Calculate packaging features ,in , Indicates the first individual products Inner The weighting coefficient of each packaging component, Indicates the first individual products Inner The status score of each packaging component. Indicates the first individual products Inner The status score of each packaging component. Represents the set of required components. Indicates an indicator function, Indicates packaging components. Indicates the first individual products Inner The weighting coefficient of each packaging component, Indicates the first individual products Inner The status score of each packaging component. Indicates the serial number;

[0014] The state score of the packaging component is generated by the GNN model by comparing and analyzing images of the packaging components of the same product before the merchant ships the goods and after the return is received. It is used to quantify the degree of damage to the packaging in the return process. The score comprehensively evaluates the structural integrity, surface condition and functionality of the component, reflecting whether it can be reused or needs to be repaired or replaced. The greater the degree of damage, the smaller the state score, and vice versa.

[0015] S103, Read the... individual products Weighting coefficients for all sampling regions and color difference value ,in , Indicates the first The difference in brightness in key areas Indicates the first Color difference in key areas Indicates the first Hue difference in key areas This represents the adjustment coefficient. Indicates the first The first item Color difference values ​​of each sampling area Indicates the number of sampling areas. Indicates the first The first item The weighting coefficients of each sampling region are determined according to... and Calculate color features ,in , Indicates the serial number.

[0016] Preferably, step S2 specifically includes the following steps:

[0017] S201, Read the... individual products The corresponding number individual users The amount, return date, and tag number of all return records are used to calculate fraud characteristics based on the amount, date, and tag number of each return. ;

[0018] S202. After setting the time window size, count the number of... individual users Number of returns within the time window and number of purchases ,according to and Calculate the return frequency characteristics ,in Read the first individual users After completing all return records, calculate the corresponding return time ratio based on the return date, purchase date, and allowed return days in the records;

[0019] S203. Calculate the return time ratio of all return records. Calculate the time ratio of the mean Then, according to and Analysis of cutoff date approach characteristics , , This indicates the total number of return records. Indicates the cutoff threshold. Indicates the first The return time of the returned items is compared to the previous one. Indicates the first The return time of the returned items is compared to the previous one. The weighting coefficients representing the time ratio. Indicates the serial number.

[0020] Preferably, the fraud feature in S201 Specifically:

[0021]

[0022] in, Indicates the time decay coefficient. Indicates the first The number of days from the current date for each return record. Indicates the first The amount of the returned item. Indicates the first Total amount of refunds per user Indicates an indicator function, Represents the Laplace smoothing constant. Indicates the first The marker bit of the return record, This indicates the total number of return records. Indicates the serial number.

[0023] Preferably, the return time ratio in S203 is specifically:

[0024]

[0025] in, Indicates the first The return time of each return record is compared to, Indicates the first The number of days between the purchase date and the return date for each record. Indicates the number of days allowed for returns. This represents the penalty coefficient for exceeding the time limit. Indicates the serial number.

[0026] Preferably, step S4 specifically includes the following steps:

[0027] S401, Read the... individual products Product categories User credibility and the current length of the review queue Afterwards, according to Analyze the risk coefficient Value coefficient and density coefficient Regarding the refund amount After normalization, the processed amount is obtained. ,in , This represents the empirical minimum value of the purchase amount. This represents the maximum empirical value of the purchase amount, using... and Calculate the loss index , ,in Indicates the control coefficient. Indicates the serial number;

[0028] According to Analyze the risk coefficient Value coefficient and density coefficient Specifically, this involves reading the category risk table, user value table, and operational load table, based on the category... Look up the corresponding risk coefficient in the category risk table. According to user credibility Query the corresponding value coefficient in the user value table. Using the current review queue length Query the corresponding density coefficient in the operational load table. ;

[0029] S402, Utilizing the Loss Index and density coefficient Calculate the processing cost index ,in , This represents the adjustment parameter, using the value coefficient. and density coefficient Analysis of the cost index of misjudgment ,in , Indicates the attenuation coefficient. Indicates the serial number;

[0030] S403, Utilizing the Loss Index Processing cost index and the cost of misjudgment index Calculate the cost factor ,in , This represents the weighting coefficient of the index. This represents the minimum cost factor. This represents the maximum value of the cost factor. Indicates the serial number.

[0031] Preferably, step S5 specifically includes the following steps:

[0032] S501, according to the... individual products Fraud probability prediction value and cost factors After determining the corresponding partition, take appropriate measures according to the partition and calculate the corresponding impact coefficient. and direction of influence ,in , , Indicates the serial number, according to The first one awaiting manual review individual products Return records, fraud probability predictions, cost factors, and all features are assigned to auditors in corresponding categories, including those for identifying fraudulent activities and assessing losses.

[0033] S502. Calculate the fraud probability and cost factor corresponding to each return record of a specified user, map each return record to a plane point, record the timestamp, and connect the points in chronological order to form a trajectory polyline.

[0034] S503. Based on the user's trajectory line statistics, the frequency, density, and interval time of points in Zone I are statistically analyzed. The frequency, density, and interval time are weighted and fused to calculate the current user's credibility, and the value coefficient is adjusted according to the credibility.

[0035] The big data-based e-commerce transaction security assessment system includes a feature extraction unit, a fraud prediction unit, a cost factor calculation unit, and a partition determination unit.

[0036] The feature extraction unit reads the measured capacity, standard capacity, status score of packaging components, and color difference value of the specified returned goods, analyzes the capacity characteristics, standard weight characteristics, packaging characteristics, and color characteristics of the goods, obtains all return records of the user corresponding to the goods, and calculates the fraud characteristics, return frequency characteristics, and deadline approaching characteristics of the current user based on the records.

[0037] After statistically analyzing the capacity, weight, packaging, and color characteristics of a product, the fraud prediction unit determines the fraud characteristics, return frequency, and deadline approach characteristics of the user corresponding to the product. After normalizing all the characteristics, it inputs them into the deployed LightGBM model and outputs the fraud probability prediction value corresponding to the current product.

[0038] The cost factor calculation unit reads the product category, user credibility, and current review queue length, and calculates the cost factor based on the product category, user credibility, current review queue length, and return records.

[0039] The partitioning unit determines partitions based on the predicted fraud probability of the goods and cost factors, takes corresponding measures according to the partition, assigns return information awaiting manual review to reviewers, and analyzes the credibility of different users.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] 1. This invention achieves accurate identification and interception of malicious return behavior by constructing a multi-dimensional feature system and intelligent analysis model. It extracts key features from multiple dimensions such as product physical characteristics and user behavior patterns, and comprehensively analyzes the risk probability of current returned products, which significantly improves the accuracy of return fraud identification, reduces the workload of manual review, and reduces the interference of misjudgments to normal users. Through a continuous learning mechanism, the model is continuously optimized to adapt to new fraud methods, thus building an efficient and intelligent return risk control barrier for merchants.

[0042] 2. This invention combines fraud probability and cost factors in decision-making, achieving a quantitative trade-off between risk and cost. This makes the decision both intelligent and aligned with actual business needs, avoiding the rigidity of simply looking at probabilities. At the same time, it regularly analyzes users' return behavior patterns and dynamically evaluates their long-term credibility. Furthermore, key processes such as feature extraction, fraud prediction, and cost factor calculation are independent of each other. This highly cohesive and loosely coupled structure significantly improves the overall robustness. When any individual module needs optimization, there is no need to reconstruct the overall structure, greatly reducing maintenance costs. Attached Figure Description

[0043] Figure 1 An overall method flowchart is provided for embodiments of the present invention;

[0044] Figure 2 A system flowchart is provided for embodiments of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example 1:

[0047] Please see Figure 1 - Figure 2 This invention provides a technical solution: a big data-based e-commerce transaction security assessment method, comprising the following steps:

[0048] S1. Extract product features: Read the measured capacity, standard capacity, packaging component status score and color difference value of the specified returned product, and analyze the product's capacity features, standard weight features, packaging features and color features;

[0049] S2. Extract user features: Read all return records of the user corresponding to the product, and calculate the fraud features, return frequency features and deadline approach features of the current user based on the records;

[0050] S3. Predict fraud probability: After statistically analyzing the capacity, weight, packaging, and color characteristics of the product, determine the fraud characteristics, return frequency characteristics, and deadline approach characteristics of the user corresponding to the product. After normalizing all characteristics, input them into the deployed LightGBM model and output the predicted fraud probability value corresponding to the current product.

[0051] S4. Determine the cost factor: Read the product category, user credibility, and current review queue length, and calculate the cost factor based on the product category, user credibility, current review queue length, and return records;

[0052] S5. Take relevant measures: Determine the zoning based on the predicted fraud probability value and cost factors of the goods, take corresponding measures according to the zoning, assign the return information to the auditors who are to be manually reviewed, and analyze the credibility of different users.

[0053] S1 specifically includes the following steps:

[0054] S101, Read the... individual products Actual capacity of returned goods and standard capacity Then, using and Calculate capacity characteristics ,in Read Actual weight of returned goods and standard weight Then, using and Calculate the standard weight characteristics ,in , Indicates the serial number;

[0055] S102, Read the... individual products Weighting coefficients for all packaging components and status score ,according to and Calculate packaging features ,in , Indicates the first individual products Inner The weighting coefficient of each packaging component, Indicates the first individual products Inner The status score of each packaging component. Indicates the first individual products Inner The status score of each packaging component. Represents the set of required components. Indicates an indicator function, Indicates packaging components. Indicates the first individual products Inner The weighting coefficient of each packaging component, Indicates the first individual products Inner The status score of each packaging component. Indicates the serial number;

[0056] The state score of the packaging component is generated by the GNN model by comparing and analyzing images of the packaging components of the same product before the merchant ships the goods and after the return is received. It is used to quantify the degree of damage to the packaging in the return process. The score comprehensively evaluates the structural integrity, surface condition and functionality of the component, reflecting whether it can be reused or needs to be repaired or replaced. The greater the degree of damage, the smaller the state score, and vice versa.

[0057] S103, Read the... individual products Weighting coefficients for all sampling regions and color difference value ,in , Indicates the first The difference in brightness in key areas Indicates the first Color difference in key areas Indicates the first Hue difference in key areas This represents the adjustment coefficient. Indicates the first The first item Color difference values ​​of each sampling area Indicates the number of sampling areas. Indicates the first The first item The weighting coefficients of each sampling region are determined according to... and Calculate color features ,in , Indicates the serial number;

[0058] S2 specifically includes the following steps:

[0059] S201, Read the... individual products The corresponding number individual users The amount, return date, and tag number of all return records are used to calculate fraud characteristics based on the amount, date, and tag number of each return. ;

[0060] S202. After setting the time window size, count the number of... individual users Number of returns within the time window and number of purchases ,according to and Calculate the return frequency characteristics ,in Read the first individual users After completing all return records, calculate the corresponding return time ratio based on the return date, purchase date, and allowed return days in the records;

[0061] S203. Calculate the return time ratio of all return records. Calculate the time ratio of the mean Then, according to and Analysis of cutoff date approach characteristics , , This indicates the total number of return records. Indicates the cutoff threshold. Indicates the first The return time of the returned items is compared to the previous one. Indicates the first The return time of the returned items is compared to the previous one. The weighting coefficients representing the time ratio. Indicates the serial number;

[0062] Fraud characteristics in S201 Specifically:

[0063]

[0064] in, Indicates the time decay coefficient. Indicates the first The number of days from the current date for each return record. Indicates the first The amount of the returned item. Indicates the first Total amount of refunds per user Indicates an indicator function, Represents the Laplace smoothing constant. Indicates the first The marker bit of the return record, This indicates the total number of return records. Indicates the serial number;

[0065] The specific return time ratio in S203 is as follows:

[0066]

[0067] in, Indicates the first The return time of each return record is compared to, Indicates the first The number of days between the purchase date and the return date for each record. Indicates the number of days allowed for returns. This represents the penalty coefficient for exceeding the time limit. Indicates the serial number.

[0068] The specific steps for deploying the LightGBM model in S3 are as follows:

[0069] S301. After reading the historical return records of each user, statistically analyze the capacity characteristics, standard weight characteristics, packaging characteristics, and color characteristics of all products in the historical return records, determine the fraud characteristics, return frequency characteristics, and deadline approximation characteristics of each user in different historical return records, and normalize all characteristics.

[0070] S302. Store all features of a single user and the return flag bits of each record as samples in a sample set, and then sort the sample set according to... The sets are randomly divided into training, testing, and validation sets.

[0071] S303. Select LightGBM as the prediction model, initialize key parameters, train the model using training set data, use Bayesian optimization to fine-tune key hyperparameters on the validation set, implement an early stopping strategy during training, terminate training when the validation set loss does not improve for 10 consecutive rounds, save the model with the best number of iterations after training, and evaluate the model performance on the test set.

[0072] S304. Establish a model iteration mechanism, and periodically retrain the model using new samples to update the model parameters;

[0073] S4 specifically includes the following steps:

[0074] S401, Read the... individual products Product categories User credibility and the current length of the review queue Afterwards, according to Analyze the risk coefficient Value coefficient and density coefficient Regarding the refund amount After normalization, the processed amount is obtained. ,in , This represents the empirical minimum value of the purchase amount. This represents the maximum empirical value of the purchase amount, using... and Calculate the loss index , ,in Indicates the control coefficient. Indicates the serial number;

[0075] according to Analyze the risk coefficient Value coefficient and density coefficient Specifically, this involves reading the category risk table, user value table, and operational load table, based on the category... Look up the corresponding risk coefficient in the category risk table. According to user credibility Query the corresponding value coefficient in the user value table. Using the current review queue length Query the corresponding density coefficient in the operational load table. ;

[0076] S402, Utilizing the Loss Index and density coefficient Calculate the processing cost index ,in , This represents the adjustment parameter, using the value coefficient. and density coefficient Analysis of the cost index of misjudgment ,in , Indicates the attenuation coefficient. Indicates the serial number;

[0077] S403, Utilizing the Loss Index Processing cost index and the cost of misjudgment index Calculate the cost factor ,in , This represents the weighting coefficient of the index. This represents the minimum cost factor. This represents the maximum value of the cost factor. Indicates the serial number;

[0078] S5 specifically includes the following steps:

[0079] S501, according to the... individual products Fraud probability prediction value and cost factors After determining the corresponding partition, take appropriate measures according to the partition and calculate the corresponding impact coefficient. and direction of influence ,in , , Indicates the serial number, according to The first one awaiting manual review individual products Return records, fraud probability predictions, cost factors, and all features are assigned to auditors in the corresponding categories, including identifying fraudulent activities and assessing losses.

[0080] S502. Calculate the fraud probability and cost factor corresponding to each return record of a specified user, map each return record to a plane point, record the timestamp, and connect the points in chronological order to form a trajectory polyline.

[0081] S503. Based on the user's trajectory line statistics, the frequency, density, and interval time of points in Zone I are statistically analyzed. The frequency, density, and interval time are weighted and fused to calculate the current user's credibility, and the value coefficient is adjusted according to the credibility.

[0082] Example 2:

[0083] The present invention also provides an e-commerce transaction security assessment system based on big data, including a feature extraction unit, a fraud prediction unit, a cost factor calculation unit, and a partition determination unit;

[0084] The feature extraction unit reads the measured capacity, standard capacity, status score of packaging components, and color difference value of the specified returned goods, analyzes the capacity characteristics, standard weight characteristics, packaging characteristics, and color characteristics of the goods, obtains all return records of the user corresponding to the goods, and calculates the fraud characteristics, return frequency characteristics, and deadline approaching characteristics of the current user based on the records.

[0085] After the fraud prediction unit statistically analyzes the capacity, weight, packaging, and color characteristics of a product, it determines the fraud characteristics, return frequency, and deadline approach characteristics of the user corresponding to the product. After normalizing all the characteristics, it inputs them into the deployed LightGBM model and outputs the fraud probability prediction value corresponding to the current product.

[0086] The cost factor calculation unit reads the product category, user credibility, and current review queue length, and calculates the cost factor based on the product category, user credibility, current review queue length, and return records.

[0087] The partitioning unit determines partitions based on the predicted fraud probability of the goods and cost factors, takes corresponding measures according to the partition, assigns return information awaiting manual review to reviewers, and analyzes the credibility of different users.

[0088] The specific partitioning is determined based on the predicted fraud probability and cost factor of the product. After setting probability and cost thresholds, if the probability threshold is greater than or equal to the predicted fraud probability and the cost factor is greater than or equal to the cost threshold, the current refund record belongs to Zone I. If the probability threshold is less than the predicted fraud probability and the cost factor is greater than or equal to the cost threshold, it belongs to Zone II. If the probability threshold is greater than or equal to the predicted fraud probability and the cost factor is less than the cost threshold, it belongs to Zone III. If the probability threshold is less than the predicted fraud probability and the cost factor is less than the cost threshold, it belongs to Zone IV. The specific handling operations for all refund records in Zone I are as follows: immediately freeze the refund, conduct the highest priority manual review (within 2 hours), and add the associated account to the monitoring list. The specific handling operations for all refund records in Zone II are as follows: normal refund process, but with a 24-hour delay in arrival, automatically mark the next 3 orders for automatic manual review, and send a return notice reminder. The specific handling operations for all refund records in Zone III are as follows: execute the fast verification process, refund according to the normal process if there are no problems, and add the user to the "low cost, high risk" observation list. The specific handling operations for all refund transactions in Zone IV are as follows: automated processing, immediate or ultra-fast refund, and no additional review.

[0089] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0090] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A big data-based e-commerce transaction security assessment method, characterized in that, The method includes the following steps: S1. Extract product features: Read the measured capacity, standard capacity, packaging component status score and color difference value of the specified returned product, and analyze the product's capacity features, standard weight features, packaging features and color features; S2. Extract user features: Read all return records of the user corresponding to the product, and calculate the fraud features, return frequency features and deadline approach features of the current user based on the records; S3. Predict fraud probability: After statistically analyzing the capacity, weight, packaging, and color characteristics of the product, determine the fraud characteristics, return frequency characteristics, and deadline approach characteristics of the user corresponding to the product. After normalizing all characteristics, input them into the deployed LightGBM model and output the predicted fraud probability value corresponding to the current product. S4. Determine the cost factor: Read the product category, user credibility, and current review queue length, and calculate the cost factor based on the product category, user credibility, current review queue length, and return records; S5. Take relevant measures: Determine the zoning based on the predicted fraud probability value and cost factors of the goods, take corresponding measures according to the zoning, assign the return information to the auditors who are to be manually reviewed, and analyze the credibility of different users.

2. The e-commerce transaction security assessment method based on big data according to claim 1, characterized in that, S1 specifically includes the following steps: S101, Read the... individual products Actual capacity of returned goods and standard capacity Then, using and Calculate capacity characteristics ,in Read Actual weight of returned goods and standard weight Then, using and Calculate the standard weight characteristics ,in , Indicates the serial number; S102, Read the... individual products Weighting coefficients for all packaging components and status score ,according to and Calculate packaging features ,in , Indicates the first individual products Inner The weighting coefficient of each packaging component, Indicates the first individual products Inner The status score of each packaging component. Indicates the first individual products Inner The status score of each packaging component. Represents the set of required components. Indicates an indicator function, Indicates packaging components. Indicates the first individual products Inner The weighting coefficient of each packaging component, Indicates the first individual products Inner The status score of each packaging component. Indicates the serial number; S103, Read the... individual products Weighting coefficients for all sampling regions and color difference value ,in , Indicates the first The difference in brightness in key areas Indicates the first Color difference in key areas Indicates the first Hue difference in key areas This represents the adjustment coefficient. Indicates the first The first item Color difference values ​​of each sampling area Indicates the number of sampling areas. Indicates the first The first item The weighting coefficients of each sampling region are determined according to... and Calculate color features ,in , Indicates the serial number.

3. The e-commerce transaction security assessment method based on big data according to claim 1, characterized in that, S2 specifically includes the following steps: S201, Read the... individual products The corresponding number individual users The amount, return date, and tag number of all return records are used to calculate fraud characteristics based on the amount, date, and tag number of each return. ; S202. After setting the time window size, count the number of... individual users Number of returns within the time window and number of purchases ,according to and Calculate the return frequency characteristics ,in Read the first individual users After completing all return records, calculate the corresponding return time ratio based on the return date, purchase date, and allowed return days in the records; S203. Calculate the return time ratio of all return records. Calculate the time ratio of the mean Then, according to and Analysis of cutoff date approach characteristics , , This indicates the total number of return records. Indicates the cutoff threshold. Indicates the first The return time of the returned items is compared to the previous one. Indicates the first The return time of the returned items is compared to the previous one. The weighting coefficients representing the time ratio. Indicates the serial number.

4. The e-commerce transaction security assessment method based on big data according to claim 2, characterized in that, Fraud features in S201 Specifically: in, Indicates the time decay coefficient. Indicates the first The number of days from the current date for each return record. Indicates the first The amount of the returned item. Indicates the first Total amount of refunds per user Indicates an indicator function, Represents the Laplace smoothing constant. Indicates the first The marker bit of the return record, This indicates the total number of return records. Indicates the serial number.

5. The e-commerce transaction security assessment method based on big data according to claim 2, characterized in that: The return time ratio mentioned in S203 is specifically as follows: in, Indicates the first The return time of each return record is compared to, Indicates the first The number of days between the purchase date and the return date for each record. Indicates the number of days allowed for returns. This represents the penalty coefficient for exceeding the time limit. Indicates the serial number.

6. The e-commerce transaction security assessment method based on big data according to claim 1, characterized in that, S4 specifically includes the following steps: S401, Read the... individual products Product categories User credibility and the current length of the review queue Afterwards, according to Analyze the risk coefficient Value coefficient and density coefficient Regarding the refund amount After normalization, the processed amount is obtained. ,in , This represents the empirical minimum value of the purchase amount. This represents the maximum empirical value of the purchase amount, using... and Calculate the loss index , ,in Indicates the control coefficient. Indicates the serial number; S402, Utilizing the Loss Index and density coefficient Calculate the processing cost index ,in , This represents the adjustment parameter, using the value coefficient. and density coefficient Analysis of the cost index of misjudgment ,in , Indicates the attenuation coefficient. Indicates the serial number; S403, Utilizing the Loss Index Processing cost index and the cost of misjudgment index Calculate the cost factor ,in , This represents the weighting coefficient of the index. This represents the minimum cost factor. This represents the maximum value of the cost factor. Indicates the serial number.

7. The e-commerce transaction security assessment method based on big data according to claim 1, characterized in that, S5 specifically includes the following steps: S501, according to the... individual products Fraud probability prediction value and cost factors After determining the corresponding partition, take appropriate measures according to the partition and calculate the corresponding impact coefficient. and direction of influence ,in , , Indicates the serial number, according to The first one awaiting manual review individual products Return records, fraud probability predictions, cost factors, and all features are assigned to auditors of the corresponding categories. S502. Calculate the fraud probability and cost factor corresponding to each return record of a specified user, map each return record to a plane point, record the timestamp, and connect the points in chronological order to form a trajectory polyline. S503. Based on the user's trajectory line statistics, the frequency, density, and interval time of points in Zone I are statistically analyzed. The frequency, density, and interval time are weighted and fused to calculate the current user's credibility, and the value coefficient is adjusted according to the credibility.

8. A big data-based e-commerce transaction security assessment system, characterized in that: The special item detection system is applicable to the big data-based e-commerce transaction security assessment method described in any one of claims 1-7, including a feature extraction unit, a fraud prediction unit, a cost factor calculation unit, and a partition determination unit; The feature extraction unit reads the measured capacity, standard capacity, status score of packaging components, and color difference value of the specified returned goods, analyzes the capacity characteristics, standard weight characteristics, packaging characteristics, and color characteristics of the goods, obtains all return records of the user corresponding to the goods, and calculates the fraud characteristics, return frequency characteristics, and deadline approaching characteristics of the current user based on the records. After statistically analyzing the capacity, weight, packaging, and color characteristics of a product, the fraud prediction unit determines the fraud characteristics, return frequency, and deadline approach characteristics of the user corresponding to the product. After normalizing all the characteristics, it inputs them into the deployed LightGBM model and outputs the fraud probability prediction value corresponding to the current product. The cost factor calculation unit reads the product category, user credibility, and current review queue length, and calculates the cost factor based on the product category, user credibility, current review queue length, and return records. The partitioning unit determines partitions based on the predicted fraud probability of the goods and cost factors, takes corresponding measures according to the partition, assigns return information awaiting manual review to reviewers, and analyzes the credibility of different users.

Images