Artificial intelligence-based return express platform information automatic distinguishing method

Abstract

The invention discloses a return express platform information automatic distinguishing method based on artificial intelligence, and particularly relates to the field of artificial intelligence, and the method comprises the steps: S1, return express platform key information extraction, S2, return consumption port acquisition, S3, corresponding relation establishment, S4, anomaly detection, S5, return express anomaly analysis, S6, risk assessment, and S7, visual display. According to the method, key information such as consignee identifiers and express bill numbers is extracted, a multi-platform data mapping relation is established, a return fraud coefficient and a return risk coefficient are constructed for two-dimensional assessment, and in combination with abnormal feature parameters, refined classification of customer risks is realized, misjudgment of a single index is avoided, the accuracy and flexibility of risk assessment are improved, and the risk assessment efficiency is improved. The overall anti-fraud capability is enhanced, and the cost and loopholes of multi-platform collaborative governance are reduced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based method for automatically distinguishing information on a return express platform. Background Art

[0002] The scale of the e-commerce industry continues to expand. As the return rate of online shopping increases, it is higher than the return rate of traditional offline shopping. The increasing number of returns has brought tremendous processing pressure to express delivery platforms. At the same time, due to the richness of online shopping platforms, return fraud often occurs. Return fraud includes returning used goods, returning items that do not match the purchased goods, etc. It is difficult for companies to accurately identify return fraud through traditional manual review methods, resulting in huge economic losses.

[0003] Traditional methods for automatically distinguishing information on return express platforms are usually based on rule engines, simple machine learning models, or basic data matching technologies. These methods include information collection, rule determination, and data matching. Information collection is used to obtain basic data related to return express as the original basis for information distinction; rule determination automatically judges and classifies the collected information based on preset business rules; and data matching is used to verify the consistency and authenticity of return information through structured data matching.

[0004] However, it still has some shortcomings in actual use. First, the platform identification is not embedded in the express information during the return process. In particular, when customers return purchases on multiple platforms, it is impossible to effectively distinguish customer information, resulting in the inability to directly link the order platform through basic information when returning across platforms, providing space for customers to exploit loopholes such as multiple returns of one item. Based on this situation, express information identification is needed, but the existing response to such fraud is mostly limited to manual identification, which achieves distinction by manually modifying the consignee identification. It is necessary to manually set rules for each platform separately. If a new platform is added or the business is adjusted, the rules need to be maintained again. The lack of automatic adaptation capabilities limits the express information identification effect and there is a hidden danger of not being able to meet the identification needs. In addition, manual identification will interfere with the platform's work efficiency to a certain extent, resulting in low return efficiency, thereby reducing customer satisfaction.

[0005] Second, after receiving the return express, there is a problem of mismatch between the recipient's identification and the platform. In this case, refund application monitoring is required. However, since customers return products after purchasing on multiple platforms, they fill in one return information when returning the products. Customer service cannot effectively distinguish and verify, causing customers to complete refund applications on multiple platforms, resulting in serious economic losses to the company. At the same time, since the return processing time limit is different on different platforms, the company needs to intervene and handle the matter afterwards, but it cannot intercept the risk in real time during the return application stage, resulting in losses in advance. Summary of the Invention

[0006] In view of this, an embodiment of the present invention provides an artificial intelligence-based method for automatically distinguishing information of a return express platform. By extracting key information of the return express platform during the customer return process and matching the return consumption port to establish a corresponding relationship, refined and flexible adjustments are made to maximize the return express platform's ability to identify customer fraud, effectively solving the problems raised in the background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] S1: Extracting key information from the return express platform: Counting the consignee identifiers on the return express platform and extracting key information corresponding to the consignee identifiers from the return express platform, where the key information includes consignee information and return information;

[0009] S2: Acquiring the return consumption port: Searching for the consumption port on the return platform corresponding to the express order number for each consignee ID in the return information corresponding to the key information, and then using the corresponding consumption port as the return consumption port for the express order number corresponding to each consignee ID;

[0010] S3: Establishing a corresponding relationship: Mapping the consignee ID on the return express platform with the return consumption port, thereby establishing a corresponding relationship between the consignee ID and the express delivery number, and judging whether there is any abnormality in the return express based on the corresponding relationship between the consignee ID and the express delivery number;

[0011] S4: Anomaly Detection: Perform anomaly detection on the return express based on the abnormal judgment of the return express, and obtain abnormal feature parameters from the return consumption port corresponding to each consignee ID based on the anomaly detection, where the abnormal feature parameters include the platform return rate, platform ID matching degree, and the number of express delivery number associations;

[0012] S5: Return Express Anomaly Analysis: Analyze the abnormal characteristic parameters of the consignee ID corresponding to each return consumption port in each time window, and obtain the return fraud coefficient and return risk coefficient of each return consumption port corresponding to the consignee ID in the current time window;

[0013] S6: Risk Assessment: Combine the return fraud coefficient and return risk coefficient of each return consumption port corresponding to the consignee ID in the current time window to evaluate the comprehensive risk assessment index corresponding to the consignee ID, and then differentiate and adjust the return express delivery based on the comprehensive risk assessment index;

[0014] S7: Visual display: Based on the risk assessment results, the differentiation of return express platforms is summarized and a fraud handling report is generated. The fraud handling report includes the return anomaly analysis results and differentiation adjustments, and is synchronized to all related platforms to form a cross-platform anti-fraud closed loop.

[0015] Technical effects and advantages of the present invention:

[0016] 1. This invention extracts key information such as consignee identification and courier tracking number to establish a multi-platform data mapping relationship. It can accurately identify abnormal behaviors such as the cross-platform upload of the same courier tracking number and the mismatch between the consignee identification and the platform. It is not limited to manual identification. On the one hand, it realizes cross-platform data linkage detection and effectively locates fraudulent behaviors that maliciously exploit loopholes in multi-platform rules. On the other hand, it can cooperate with the customer service department across departments through artificial intelligence to distinguish returns from different platforms by modifying the consignee information, effectively improving processing efficiency, faster identifying abnormal returns, and reducing company losses.

[0017] 2. The present invention conducts a two-dimensional assessment by constructing a return fraud coefficient and a return risk coefficient, and combines multi-source abnormal characteristic parameters such as the platform return rate, identification matching degree, and the number of express order number associations to achieve refined classification of customer risks, avoid misjudgment of a single indicator, and improve the accuracy and flexibility of risk assessment. By generating a fraud processing report and synchronizing it to the return express platform, it ensures that each platform shares high-risk user tags in real time, prevents fraudulent users from transferring across platforms, enhances overall anti-fraud capabilities, and reduces the cost and loopholes of multi-platform collaborative governance. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the overall structure of the present invention.

[0019] Figure 2 This is a flow chart of the platform return rate acquisition method of the present invention.

[0020] Figure 3 This is a schematic diagram of the distinguishing, adjusting and judging structure of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] As attached Figure 1-2 The method for automatically distinguishing information on a return express platform based on artificial intelligence includes:

[0023] S1: Extraction of key information of the return express platform: Count the consignee identifications existing on the return express platform. The consignee identification mentioned here is determined by the consignee's purchasing software. For example, the purchasing software includes but is not limited to Douyin, Tmall, and Kuaishou. If the consignee purchases goods on Douyin, the consignee will be marked as the Douyin return group, and the key information corresponding to the consignee identification will be extracted from the return express platform, where the key information includes consignee information and return information, and the consignee information includes the consignee identity hash value and historical return records. The return information includes the express order number, return reason and product price.

[0024] In this embodiment, it should be specifically noted that the user's historical return records are used to identify high-risk users, such as multiple cross-platform returns in a short period of time; the reason for the return is the return label when the user returns the product on the purchase platform. For example, the return label can be dislike, quality issues, damaged goods, etc.

[0025] It needs to be explained that the key information corresponding to the consignee identification is extracted, and the consignee identity hash value is used as the consignee information because the consignee identity hash value can represent the unique identification of the consignee identity. If the consignee ID is relied upon to identify the return express platform information, in this case, when customers use different accounts to place orders across platforms, they may be forged as orders placed by different customers, resulting in duplicate refund amounts and increasing the company's losses.

[0026] S2: Obtaining the return consumption port: Extract the consignee identity hash value and express delivery number from the return information corresponding to the key information. At this time, connect to the return platform according to the consignee name and express delivery number corresponding to each return express. The return platform is a complex of various return express delivery platforms, specifically the complex corresponding to Douyin, Tmall, Kuaishou, etc., and search for the consumption port of the express delivery number corresponding to each consignee identification on the return platform. The consumption port is the order return platform, and then the corresponding consumption port is used as the return consumption port of the express delivery number corresponding to each consignee identification.

[0027] S3: Establishing a corresponding relationship: Map the consignee ID on the return express platform with the return consumption port, thereby establishing a corresponding relationship between the consignee ID and the express delivery number, and determine whether there is any abnormality in the return express based on the corresponding relationship between the consignee ID and the express delivery number.

[0028] In this embodiment, it is necessary to specifically explain the following process of establishing the correspondence between the consignee identifier and the express delivery number:

[0029] A1: The consignee IDs on the return courier platform are numbered in a set order, labeled 1, 2, ..., n, where the set order is the order of the consignee's return time;

[0030] A2: Keyword recognition is performed based on the return consumption port and the consignee ID. The order return platform and the purchase software are matched to form a mapping group of the return consumption port and the consignee ID.

[0031] A3: Extract the express delivery order number through the return consumption port, and build a corresponding relationship between the return consumption port and the express delivery order number based on the return consumption port corresponding to the consignee ID in each mapping group.

[0032] It should be further explained that since the express delivery number generated when purchasing goods is unique, when establishing the correspondence between the consignee identification and the express delivery number, if there is a unique correspondence between the express delivery number and the consignee identification, the express delivery number is judged to be normal and the express delivery is returned normally through the return express platform. If there is a correspondence between the express delivery number and multiple consignee identifications, the express delivery number is judged to be abnormal.

[0033] S4: Anomaly detection: Perform anomaly detection on the return express based on the abnormal judgment of the return express, and obtain abnormal feature parameters from the return consumption port corresponding to each consignee ID based on the anomaly detection, where the abnormal feature parameters include the platform return rate, platform ID matching degree and the number of express order number associations.

[0034] It needs to be explained that due to the immediacy of malicious returns and the lag of logistics tracks, a time window needs to be set before anomaly detection. The express arrival time is set as the starting time, and the time interval after the starting time is taken as the first time window. The time windows are marked as 1, 2, ..., j, ..., k in sequence. For example, the time interval can be 24 hours, 12 hours or 6 hours.

[0035] In this embodiment, it should be specifically explained that the specific method for obtaining the platform return rate includes the following steps:

[0036] Obtain the return records of each return courier platform corresponding to each time window through each return consumption port, and extract the recipient's historical return times from the return records;

[0037] Extract the consignee identity hash value from the return records of each return express platform, and compare the consignee identity hash values ​​corresponding to the return records on each return express platform to determine whether there is an identical consignee identity hash value. If so, record the consignee with the same consignee identity hash value as a frequent return customer, and record the consignee with a different consignee identity hash value as a new return customer, and summarize the return records corresponding to the frequent return customers;

[0038] Count the total number of returns and active time of frequent return customers. The active time is the length of time the frequent return customers use the product within each time window. Compare the total number of returns with the active time to obtain the frequent return customer return rate.

[0039] Count the number of returns and purchases by new returning customers across multiple platforms within each time window, and compare the number of returns and purchases across multiple platforms to obtain the return rate for new returning customers;

[0040] The platform return rate is obtained by taking the weighted average of the return rate of frequent returning customers and the return rate of new returning customers. The greater the return rate of frequent returning customers and the return rate of new returning customers, the greater the platform return rate, and the higher the abnormal risk of the return express platform.

[0041] The specific method for obtaining the platform identification matching degree includes the following steps:

[0042] Build a platform rule library for each platform, including the platform name, the courier number prefix rule, and the consignee identifier regular expression. For example, when the consignee identifier is the Tmall delivery group and the return express platform is Tmall, the courier number prefix rule can be TM, and the consignee identifier regular expression can be a combination of the Tmall delivery group and the digital number. In this case, the platform rule can specifically be a combination of Tmall, TM, and the corresponding consignee identifier regular expression, and it is stored in the return rule database;

[0043] Obtain the logistics waybill image through the return express platform, and perform noise reduction and binarization on the logistics waybill image, thereby converting the image into black and white and highlighting the text area;

[0044] Identify key areas based on the logistics bill image, including the consignee ID, delivery address, and return courier platform;

[0045] Match key areas using the platform's rule base, including prefix rule matching and consignee ID regular matching, to obtain matching scores.

[0046] The platform identification matching degree is calculated based on the matching score of prefix rule matching and consignee identification regular matching, which is specifically expressed as:

[0047]

[0048] Among them, CD represents the platform identification matching degree, Qt and Ql represent the prefix rule matching score and consignee identification regular matching score corresponding to the return express platform, respectively.

[0049] The return rule database is used to store the platform rules corresponding to the platform rule base, the platform identification matching range corresponding to various platform matching coefficients, the commodity price difference coefficient of return express relative to the commodity price, and the compensation factors corresponding to the return fraud coefficient and return risk coefficient.

[0050] It should be further explained that the prefix rule matching score is used to verify whether the prefix of the express delivery number complies with the platform rules. Specifically, the keyword recognition technology can be used to judge the match between the prefix of the express delivery number and the return express delivery platform. If the prefix of the express delivery number completely matches the return express delivery platform, the prefix rule matching score is 1. Otherwise, the string difference is calculated by editing distance, and the similarity is used as the prefix rule matching score. The similarity is specifically expressed as: similarity = 1-edit distance / max (rule length, prefix length); the consignee identification regular matching score is used to determine whether the consignee identification regular expression complies with the platform rules. Specifically, the consignee identification regular expression can be matched with the regular expression of the corresponding platform rules of each platform to calculate the matching length, thereby comparing the matching length with the regular expression to obtain the text similarity, and the text similarity is used as the consignee identification regular matching score.

[0051] The specific method for obtaining the number of platforms associated with an express delivery number is to crawl the number of different platforms associated with it within 24 hours from the return consumption port of each express delivery number corresponding to each consignee identifier through a global index. For example, if the same express delivery number appears in the return application of Douyin, Tmall, and Xiaohongshu at the same time, the number of associated platforms is 3.

[0052] It needs to be explained that cross-platform fraud involving multiple returns for one product is usually time-sensitive. In order to avoid exposure of logistics information, customers will submit return applications on multiple platforms in a short period of time. The time window can accurately capture such abnormal behavior and avoid the dilution of fraud characteristics due to a long time span. At the same time, there is usually a delay of several hours from the time logistics information is collected to the time it is updated in the system. If the time window is too long, customers may take advantage of the loophole of unsynchronized logistics status to complete refunds on multiple platforms, thereby increasing the company's losses.

[0053] S5: Abnormal analysis of return express delivery: Analyze the abnormal characteristic parameters of each return consumption port corresponding to the consignee ID in each time window, and obtain the return fraud coefficient and return risk coefficient of each return consumption port corresponding to the consignee ID in the current time window.

[0054] In this embodiment, it should be specifically explained that the return fraud coefficient is obtained by analyzing the platform return rate and the matching degree of the platform identifier in the abnormal characteristic parameters corresponding to each return consumption port in each time window of the consignee identifier. The specific analysis method is as follows:

[0055] Extract the platform identification matching degree from the abnormal characteristic parameters of the consignee identification corresponding to each return consumption port in each time window, and match it with the platform identification matching degree range corresponding to various platform matching coefficients in the return rule database, thereby obtaining the platform matching coefficient of the consignee identification corresponding to the return consumption port in each time window, where the platform matching coefficient is directly proportional to the platform identification matching degree;

[0056] The return fraud coefficient of the consignee ID in the current time window is calculated based on the platform return rate and platform matching coefficient of each return consumption port in each time window corresponding to the consignee ID. It is specifically expressed as:

[0057]

[0058] Among them, Rd identifies the return fraud coefficient of the consignee ID in the current time window, Ed and Td represent the platform return rate and platform matching coefficient of each return consumption port corresponding to the consignee ID in each time window, respectively. The higher the platform return rate and the smaller the platform matching coefficient, the higher the probability of customer fraud during the return process.

[0059] It should be further explained that the return risk coefficient is obtained by analyzing the number of express delivery numbers associated with the abnormal characteristic parameters of each return consumption port in each time window corresponding to the consignee ID. The specific analysis method is as follows:

[0060] Extract return information from key information corresponding to the consignee ID;

[0061] The product price is extracted from the return information corresponding to the key information of the consignee ID, and then compared with the product price difference coefficient of the return courier relative to the product price in the return rules database to obtain the product price difference coefficient of each return consumption port corresponding to the consignee ID relative to the product price. The product price difference coefficient of the return courier relative to the product price in the return rules database is determined based on the difference between the consumption level in the customer's historical consumption record and the current consumption level of the return courier;

[0062] The number of platforms associated with the express order number is extracted from the abnormal characteristic parameters of each return consumption port in each time window corresponding to the consignee ID. The number of platforms associated with the express order number, the commodity price, and the commodity price difference coefficient are used to establish a profit-risk prediction mathematical model. The profit-risk prediction coefficient is obtained through analysis, which is specifically expressed as:

[0063]

[0064] Where Rt represents the return-risk prediction coefficient, Nr, Pr, and Yr represent the number of platforms associated with the express order number, the product price, and the product price difference coefficient, respectively; Pr0 represents the average price of the corresponding product on each platform; and e represents a natural constant.

[0065] The profit loss is obtained by multiplying the profit risk prediction coefficient by the commodity price, and the average profit loss of each return consumption port corresponding to the consignee ID in the current time window is calculated. The average profit loss is divided by the product of the commodity price and the number of platforms associated with the express order number to obtain the return risk coefficient of the consignee ID in the current time window. The greater the average profit loss, the more platforms associated with the express order number, and the greater the return risk coefficient, indicating that the return risk is higher at this time.

[0066] It needs to be explained that a large number of platforms associated with express delivery numbers may mean that users return goods on multiple platforms, which may lead to the risk of multiple returns for the same product, increasing the platform's losses. Therefore, this variable has a positive impact on the risk coefficient. If the price of the product is high, the loss caused by the return will be greater. The product price difference coefficient may refer to the price difference of the product on different platforms. The greater the difference, the more likely users are to use the price difference for arbitrage, such as purchasing on a high-priced platform and returning on a low-priced platform. Therefore, the more platforms associated with the express delivery number, the higher the product price, and the larger the product price difference coefficient, the greater the profit risk.

[0067] S6: Risk assessment: Combine the return fraud coefficient and return risk coefficient of each return consumption port corresponding to the consignee identification in the current time window to evaluate the comprehensive risk assessment index corresponding to the consignee identification, and then differentiate and adjust the return express based on the comprehensive risk assessment index.

[0068] In this embodiment, it should be specifically noted that the comprehensive risk assessment index is obtained by analyzing the return fraud coefficient and the return risk coefficient within the current time window, and is specifically expressed as:

[0069]

[0070] Among them, RB represents the comprehensive risk assessment index corresponding to the consignee identifier, Rd and Rt represent the return fraud coefficient and return risk coefficient of each return consumption port corresponding to the consignee identifier in the current time window, respectively, n represents the total number of consignee identifiers, a and b represent the compensation factors corresponding to the return fraud coefficient and return risk coefficient, respectively. The compensation factors corresponding to the return fraud coefficient and return risk coefficient are obtained from the return rule database. Based on historical data, a mapping set of historically detected return fraud coefficients and return risk coefficients and compensation factors is established to obtain the compensation factors corresponding to the current return fraud coefficient and return risk coefficient.

[0071] The present invention integrates the return fraud coefficient and return risk coefficient of the consignee identification into the comprehensive risk assessment index, making the evaluation indicators of the comprehensive risk assessment index more comprehensive and conducive to improving the accuracy of the evaluation results.

[0072] It should be further explained that before implementing differentiation adjustment, differentiation adjustment judgment needs to be made, see Figure 3 The specific operation of the differentiation and adjustment judgment shown is as follows: compare the comprehensive risk assessment index with the set risk threshold. If the comprehensive risk assessment index is greater than the set risk threshold, it is judged that differentiation and adjustment are required; otherwise, it is judged that differentiation and adjustment are not required.

[0073] Furthermore, the differentiated adjustment is implemented as follows: when the return fraud coefficient reaches the set risk threshold but the return risk coefficient does not reach the set risk threshold, this means that there is a high risk of customer fraud, while the return loss is low at this time. In this case, the adjustment level is determined to be low risk, and the consignee identity hash value is extracted from the key information corresponding to the consignee ID, and the consignee is marked as a high-risk user. At the same time, the consignee ID is automatically queried and the logistics waybill ownership is verified. If it is detected that the consignee ID does not match the target platform, the consignee information is changed, and other express refunds that do not match the target platform are intercepted;

[0074] When both the return fraud coefficient and the return risk coefficient reach the set risk threshold, the adjustment level is determined to be high risk, and the refund is stopped and all express deliveries are intercepted. After the interception is triggered, manual review will be carried out.

[0075] S7: Visual display: Based on the risk assessment results, the differentiation of return express platforms is summarized and a fraud handling report is generated. The fraud handling report includes the return anomaly analysis results and differentiation adjustments, and is synchronized to all related platforms to form a cross-platform anti-fraud closed loop.

[0076] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0077] Finally: 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 in the scope of protection of the present invention.

Claims

1. The method for automatically distinguishing information on the return express platform based on artificial intelligence is characterized by: include: S1: Extracting key information from the return express platform: Counting the consignee identifiers on the return express platform and extracting key information corresponding to the consignee identifiers from the return express platform, where the key information includes consignee information and return information; S2: Acquiring the return consumption port: Searching for the consumption port on the return platform corresponding to the express delivery number of each consignee ID in the return information corresponding to the key information, and then using the corresponding consumption port as the return consumption port for the express delivery number corresponding to each consignee ID; S3: Establishing a corresponding relationship: Mapping the consignee ID on the return express platform with the return consumption port, thereby establishing a corresponding relationship between the consignee ID and the express delivery number, and judging whether there is any abnormality in the return express based on the corresponding relationship between the consignee ID and the express delivery number; S4: Anomaly Detection: Perform anomaly detection on the return express based on the abnormal judgment of the return express, and obtain abnormal feature parameters from the return consumption port corresponding to each consignee ID based on the anomaly detection, where the abnormal feature parameters include the platform return rate, platform ID matching degree, and the number of express delivery number associations; S5: Return Express Anomaly Analysis: Analyze the abnormal characteristic parameters of each return consumption port corresponding to the consignee ID in each time window, and obtain the return fraud coefficient and return risk coefficient of each return consumption port corresponding to the consignee ID in the current time window; S6: Risk Assessment: Combine the return fraud coefficient and return risk coefficient of each return consumption port corresponding to the consignee ID in the current time window to evaluate the comprehensive risk assessment index corresponding to the consignee ID, and then differentiate and adjust the return express delivery based on the comprehensive risk assessment index; S7: Visual display: Based on the risk assessment results, the differentiation of return express platforms is summarized and a fraud handling report is generated. The fraud handling report includes the return anomaly analysis results and differentiation adjustments, and is synchronized to all related platforms to form a cross-platform anti-fraud closed loop.

2. The method for automatically distinguishing information on a return express platform based on artificial intelligence according to claim 1, characterized in that: The specific method for obtaining the platform return rate includes the following steps: Obtain the return records of each return courier platform corresponding to each time window through each return consumption port, and extract the recipient's historical return times from the return records; Extract the consignee identity hash value from the return records of each return express platform, and compare the consignee identity hash values ​​corresponding to the return records on each return express platform to determine whether there is an identical consignee identity hash value. If so, record the consignee with the same consignee identity hash value as a frequent return customer, and record the consignee with a different consignee identity hash value as a new return customer, and summarize the return records corresponding to the frequent return customers; Count the total number of returns and active time of frequent return customers. The active time is the length of time a frequent return customer uses the product within each time window. Compare the total number of returns with the active time to obtain the frequent return customer return rate. Count the number of returns and purchases by new returning customers across multiple platforms within each time window, and compare the number of returns and purchases across multiple platforms to obtain the return rate for new returning customers; The platform return rate is obtained by taking the weighted average of the return rate of frequent returning customers and the return rate of new returning customers.

3. The method for automatically distinguishing information on a return express platform based on artificial intelligence according to claim 1, characterized in that: The specific method for obtaining the platform identification matching degree includes the following steps: Build a platform rule library for each platform, including the platform name, express number prefix rules, and consignee identification regular expressions, and store them in the return rule database; Obtain the logistics waybill image through the return express platform, and perform noise reduction and binarization on the logistics waybill image, thereby converting the image into black and white and highlighting the text area; Identify key areas based on the logistics bill image, including the consignee ID, delivery address, and return courier platform; Match key areas using the platform's rule base, including prefix rule matching and consignee ID regular matching, to obtain matching scores. The platform identification matching degree is calculated based on the matching score of prefix rule matching and consignee identification regular matching, which is specifically expressed as: Among them, CD represents the platform identification matching degree, Qt and Ql represent the prefix rule matching score and consignee identification regular matching score corresponding to the return express platform, respectively.

4. The method for automatically distinguishing information on a return express platform based on artificial intelligence according to claim 3, characterized in that: The prefix rule matching score is used to verify whether the courier number prefix complies with the platform rules. The keyword recognition technology is used to match the courier number prefix with the return courier platform. If the courier number prefix completely matches the return courier platform, the prefix rule matching score is 1. Otherwise, the string difference is calculated using the edit distance, and the similarity is used as the prefix rule matching score. The consignee identification regular expression matching score is used to determine whether the consignee identification regular expression complies with the platform rules. The matching length of the consignee identification regular expression is calculated with the regular expression of the corresponding platform rules of each platform. The matching length is then compared with the regular expression to obtain the text similarity, and the text similarity is used as the consignee identification regular expression matching score.

5. The method for automatically distinguishing information on a return express platform based on artificial intelligence according to claim 1, characterized in that: The return fraud coefficient is obtained by analyzing the platform return rate and platform identification matching degree in the abnormal characteristic parameters of each return consumption port corresponding to the consignee identification in each time window. The specific analysis method is as follows: Extract the platform identification matching degree from the abnormal characteristic parameters of the consignee identification corresponding to each return consumption port in each time window, and match it with the platform identification matching degree range corresponding to various platform matching coefficients in the return rule database, thereby obtaining the platform matching coefficient of the consignee identification corresponding to the return consumption port in each time window, where the platform matching coefficient is directly proportional to the platform identification matching degree; The return fraud coefficient of the consignee ID in the current time window is calculated based on the platform return rate and platform matching coefficient of each return consumption port in each time window corresponding to the consignee ID. It is specifically expressed as: Among them, Rd represents the return fraud coefficient of the consignee ID in the current time window, and Ed and Td represent the platform return rate and platform matching coefficient of each return consumption port corresponding to the consignee ID in each time window.

6. The method for automatically distinguishing information on a return express platform based on artificial intelligence according to claim 1, characterized in that: The return risk coefficient is obtained by analyzing the number of express delivery numbers associated with the abnormal characteristic parameters of each return consumption port in each time window corresponding to the consignee identifier. The specific analysis method is as follows: Extract return information from key information corresponding to the consignee ID; The product price is extracted from the return information corresponding to the key information of the consignee ID, and then compared with the product price difference coefficient of the return courier relative to the product price in the return rules database to obtain the product price difference coefficient of each return consumption port corresponding to the consignee ID relative to the product price. The product price difference coefficient of the return courier relative to the product price in the return rules database is determined based on the difference between the consumption level in the customer's historical consumption record and the current consumption level of the return courier; The number of platforms associated with the express order number is extracted from the abnormal characteristic parameters of each return consumption port in each time window corresponding to the consignee ID. The number of platforms associated with the express order number, the product price, and the product price difference coefficient are used to establish a profit-risk prediction mathematical model, thereby analyzing and obtaining the profit-risk prediction coefficient. Based on the profit-risk prediction coefficient, the average profit loss of each return consumption port corresponding to the consignee ID in the current time window is calculated, and the average profit loss is divided by the product of the commodity price and the number of platforms associated with the express order number to obtain the return risk coefficient of the consignee ID in the current time window.

7. The method for automatically distinguishing information on a return express platform based on artificial intelligence according to claim 1, characterized in that: The comprehensive risk assessment index is obtained by analyzing the return fraud coefficient and the return risk coefficient within the current time window, and is specifically expressed as: Where RB represents the comprehensive risk assessment index corresponding to the consignee ID, Rd and Rt represent the return fraud coefficient and return risk coefficient of each return consumption port corresponding to the consignee ID in the current time window, n represents the total number of consignee IDs, a and b represent the compensation factors corresponding to the return fraud coefficient and return risk coefficient, respectively.

8. The method for automatically distinguishing information on a return express platform based on artificial intelligence according to claim 1, characterized in that: Before the implementation of the differentiation adjustment, a differentiation adjustment judgment needs to be made. The specific operation is as follows: compare the comprehensive risk assessment index with the set risk threshold. If the comprehensive risk assessment index is greater than the set risk threshold, it is judged that differentiation adjustment is required; otherwise, it is judged that differentiation adjustment is not required.

9. The method for automatically distinguishing information on a return express platform based on artificial intelligence according to claim 1, characterized in that: The differentiation adjustment is implemented as follows: when the return fraud coefficient reaches the set risk threshold but the return risk coefficient does not reach the set risk threshold, the adjustment level is determined to be low risk; When both the return fraud coefficient and the return risk coefficient reach the set risk threshold, the adjustment level is determined to be high risk, and the refund is stopped and all express deliveries are intercepted. After the interception is triggered, manual review will be carried out.

Images