Post-payment intelligent payment method and system based on pre-authorization supervision

Abstract

The invention discloses a post-payment intelligent payment method and system based on pre-authorization supervision, and aims to improve payment security and user experience. The method comprises the following steps: firstly, collecting historical purchase behavior data and operation data of a target user, and constructing operation fingerprint features of the user under different purchase behaviors; through clustering analysis, historical purchase operation diversity indexes of the user in various purchase behaviors are calculated and determined, and behavior mode stability of the user is described. And then, real-time purchase operation data of the user is acquired, a real-time diversity index is calculated, and the current purchase risk level of the user is evaluated in combination with historical data. And according to the risk result, pre-authorization supervision control is carried out on the transaction between the user and the merchant account, so that a risk-controllable post-payment strategy is realized. The method can dynamically identify abnormal transaction behaviors, enhances the risk prevention capability of a payment system, and is suitable for various online and offline intelligent payment scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of postpaid, in particular to a postpaid intelligent payment method and system based on pre-authorization supervision. BACKGROUND

[0002] With the rapid development of mobile payment and e-commerce, users' requirements for payment convenience, fund security and credit protection are continuously increasing, and the "enjoy first and pay later" payment method has gradually been favored by the market. Most of the existing traditional payment processes take the form of completing fund transfer immediately after the user places an order, lacking effective fund supervision and transaction confirmation mechanisms. Once the goods are not delivered or the service is disputed, it is difficult for the user to seek redress and for the platform to hold the business accountable, which can easily lead to commercial fraud, transaction disputes, and even business running away with money, seriously affecting the stability and credibility of the payment system.

[0003] The current market "enjoy first and pay later" products such as Alipay sesame points and WeChat payment points only use user credit scores as the basis for risk judgment, without introducing a real and controllable fund locking mechanism. This type of credit guarantee method has a certain flexibility in small transactions, but lacks sufficient protection for businesses in medium and high value transactions, and cannot effectively resist malicious fraud.

[0004] With the popularization and application of the central bank digital RMB (M0), it has the characteristics of unique account, controllable funds, specified use, and full traceability, providing a technical basis for realizing fund management and intelligent payment under strong supervision. If combined with AI intelligent risk control model, user behavior analysis and fund supervision mechanism, a safe, controllable and accountable postpaid payment system can be built based on the central bank digital RMB M0 account, which will help to completely improve the structural defects of traditional payment and credit mechanisms, improve the platform's credibility and user trust, and meet the risk control and compliance requirements in complex transaction scenarios. SUMMARY

[0005] To solve at least one of the above technical problems, the present application provides a postpaid intelligent payment method and system based on pre-authorization supervision.

[0006] The first aspect of the present application provides a postpaid intelligent payment method based on pre-authorization supervision, comprising: obtaining historical purchase behavior data and historical purchase operation data of a target user, and constructing purchase operation fingerprints of different purchase behaviors of the target user; performing clustering operation on the purchase operation fingerprints to determine the historical purchase operation diversity index of the target user for different purchase behaviors; obtaining the real-time purchase operation diversity index of the target user, and determining the real-time purchase risk of the target user according to the historical purchase operation diversity index and the real-time purchase operation diversity index; According to the real-time purchase risk, the target user and the corresponding merchant account are pre-authorized supervision, and the post-payment strategy of the target user is constructed.

[0007] In the scheme, the historical purchase behavior data and the historical purchase operation data of the target user are obtained, and the purchase operation fingerprints of different purchase behaviors of the target user are constructed, specifically: The historical purchase behavior data of the target user is obtained, and the historical purchase behavior data includes purchase commodity type, purchase price, and purchase frequency data; The historical purchase operation data of the target user for different purchase behaviors is obtained, and the historical purchase operation data includes the product search and selection operation within a preset time period before the user completes the product purchase, the click and browse operation of each information display board of the product, the customer service inquiry operation, the collection and purchase operation, and the payment operation, and the operation time length information of each purchase operation type is obtained; Each purchase operation type of the historical purchase operation data is sorted according to the operation timestamp, and a purchase operation sequence is constructed, and the purchase operation sequence is feature extracted, including the duration of each purchase operation type, the interval time between adjacent operations, the operation type conversion frequency, the operation repetition number, and the page scrolling speed; The purchase operation sequence features construct a feature vector, an hidden Markov model is introduced to model the feature vector, the purchase operation sequence corresponding to each purchase behavior is taken as an observation sequence, and the probability distribution of the observation sequence is calculated through a forward-backward algorithm; According to the probability distribution, a Viterbi algorithm is used to decode the hidden state sequence of the purchase operation sequence, the hidden state sequence is fused with the corresponding feature vector, and the fused feature vector is subjected to principal component analysis dimension reduction processing to obtain a dimension-reduced feature vector; Based on hash coding, a hash value of each dimension-reduced feature vector is constructed, the hash values are constructed into a hash fingerprint set, and the purchase operation fingerprints of different purchase behaviors of the target user are obtained.

[0008] In the scheme, the purchase operation fingerprints are clustered, and the historical purchase operation diversity index of the target user for different purchase behaviors is determined, specifically: The purchase operation fingerprints, i.e., the hash fingerprint set, are normalized and pretreated, the Hamming distance between each purchase operation fingerprint after normalization is calculated, and a distance matrix between each purchase operation fingerprint is constructed according to the Hamming distance; A density clustering algorithm is introduced, the initial neighborhood search radius ε is calculated according to the standard deviation of all purchase operation fingerprints in each dimension and the total number of fingerprint points, and the initial minimum number of points MinPts in the neighborhood is estimated based on the distance distribution of the distance matrix; Traverse all purchase operation fingerprint points, select a point not yet visited as a core point, find all fingerprint points contained in the core point epsilon neighborhood through neighborhood search radius epsilon to form a temporary subset; If the number of fingerprint points contained in the temporary subset is greater than or equal to the minimum number of points MinPts in the neighborhood, a new cluster is created, the core point is added to the new cluster, and the temporary subset is recursively expanded for each fingerprint point in the temporary subset to check whether the number of points in the epsilon neighborhood of each adjacent point in the temporary subset reaches MinPts, if so, continue to mark the adjacent point as a new core point and expand its neighborhood range until no new adjacent core point is found, thereby forming a complete cluster; If the number of points in the core point epsilon neighborhood is less than MinPts, the core point is marked as a boundary point, after the traversal of all core points and the generation of clusters, all boundary points are accessed, and the distance from each boundary point to the core points in each generated cluster is checked in turn, if there is a core point with a distance less than or equal to the neighborhood search radius epsilon from the boundary point, and the core point has been classified into a cluster, the boundary point is merged into the cluster, and the purchase operation fingerprint clustering result is obtained; Obtain the total number K of clusters in the purchase operation fingerprint clustering result, calculate the cluster center of each cluster, and calculate the internal dispersion of each cluster according to the average distance between all purchase operation fingerprints in the cluster and the cluster center; Determine the historical purchase diversity index of the purchase operation fingerprint in the cluster according to the product of the total number K of clusters and the weighted sum of the internal dispersion of each cluster, and determine the purchase behavior of each cluster to obtain the historical purchase operation diversity index of the target user for different purchase behaviors.

[0009] In this scheme, the real-time purchase risk of the target user is determined according to the historical purchase operation diversity index and the real-time purchase operation diversity index, specifically: Obtain the real-time purchase behavior data of the target user before confirming the commodity order and before the payment of the commodity, and the real-time purchase operation data in a preset time period before the payment of the commodity; Construct a real-time purchase operation sequence according to the real-time purchase operation data, perform feature extraction on the real-time purchase operation sequence, and construct a real-time purchase operation feature vector; Compare the real-time purchase operation feature vector with the reduced feature vector in terms of similarity, determine the belonging probability distribution of the current real-time purchase operation feature vector in each cluster by calculating the distance between the real-time feature vector and the cluster center of each purchase operation fingerprint, and According to the belonging probability distribution and the historical purchase operation diversity index corresponding to each cluster, the real-time purchase operation diversity index of the current purchase operation is constructed by weighted fusion; Calculate the degree of deviation between the real-time purchase operation diversity index and the historical purchase operation diversity index, and determine the real-time purchase risk of the target user according to the degree of deviation.

[0010] In this scheme, the real-time purchase risk of the target user is monitored, and the post-payment strategy of the target user is constructed, specifically: According to the real-time purchase risk, if the real-time purchase risk is greater than a preset risk value, a post-payment pre-authorization instruction is initiated to the user's payment account, and when the post-payment pre-authorization instruction of the user is received, the supervision level of the user's payment account is determined according to the real-time purchase risk; According to the supervision level, the frozen proportion of the commodity payment amount of the user account and the frozen duration are determined, and the post-payment pre-authorization supervision parameter is obtained; According to the post-payment pre-authorization supervision parameter, the user account is pre-authorized, the smart contract is initialized according to the post-payment pre-authorization supervision parameter, the smart contract includes user identity, merchant account information, commodity amount details, commodity description and pre-authorization supervision parameter, and the smart contract is bound with a preset account; According to the smart contract, the pre-authorization operation is executed, the commodity payment amount of the user is transferred from the user account to the preset account according to the frozen proportion to freeze, a unique pre-authorization voucher is generated, and the frozen state and frozen time are recorded, and the smart contract and the unique pre-authorization voucher, the frozen state and the frozen time are synchronized to the preset insurance account; Real-time monitoring of merchant delivery and user receipt state data, including logistics state, merchant delivery voucher, and user signature state; According to the merchant delivery and user receipt state data, the commodity delivery of the user is determined, if the commodity has been delivered to the user, the payment instruction is executed based on the smart contract and the supervision level, the frozen fund is transferred to the merchant account, if the payment is abnormal, the frozen amount is transferred to the preset insurance account, and the merchant is compensated according to the preset insurance account; If the user does not confirm receipt or raise objections within a preset time, the responsibility is determined according to the platform, and the delivery is performed according to the determination result to obtain the post-payment strategy of the target user.

[0011] In this scheme, the real-time purchase risk of the target user is monitored, and the post-payment strategy of the target user is constructed, specifically: When the real-time purchase risk of the user account exceeds the preset risk value for a preset number of times, the historical pre-authorization failure record and the corresponding risk fluctuation data of the user account are obtained, and the deviation coefficient of the historical risk fluctuation variance and the current real-time purchase risk is calculated. If the deviation coefficient exceeds the dynamic threshold, an interaction request containing a behavior verification question is sent to the user terminal, the user's operation response sequence to the verification question is obtained, and the operation interval time, input correction times and trajectory features in the response sequence are extracted; When the operation interval time is less than the historical average response time and the input correction times are lower than the abnormal threshold, it is determined as a manual operation, and the pre-authorization instruction is continued to be executed, otherwise the biometric feature verification is started, and the real-time face image and voiceprint feature of the user are obtained; The real-time biometric feature is matched with the pre-stored template, when the matching similarity is lower than the first threshold but higher than the second threshold, the location of the payment account associated device is checked, if the deviation distance of the device location from the commonly used geographic fence exceeds the preset range, the account is frozen and a security alarm is sent; The real-time purchase risk value is re-evaluated according to the operation response sequence, biometric feature verification and location check, if the risk value decreases by more than the drawdown threshold, the freezing proportion is reduced and the freezing duration is shortened, otherwise the original pre-authorization supervision parameters are maintained.

[0012] The second aspect of the application also provides a postpaid intelligent payment system based on pre-authorization supervision, which comprises a memory and a processor, wherein the memory comprises a postpaid intelligent payment method program based on pre-authorization supervision, and the postpaid intelligent payment method program based on pre-authorization supervision is executed by the processor to realize the following steps: Obtain the historical purchase behavior data and historical purchase operation data of the target user, and construct the purchase operation fingerprints of different purchase behaviors of the target user; Cluster the purchase operation fingerprints to determine the historical purchase operation diversity index of the target user for different purchase behaviors; Obtain the real-time purchase operation diversity index of the target user, and determine the real-time purchase risk of the target user according to the historical purchase operation diversity index and the real-time purchase operation diversity index; Pre-authorization supervision is performed on the target user and the corresponding merchant account according to the real-time purchase risk, and a postpaid strategy of the target user is constructed.

[0013] The application discloses a post-paid intelligent payment method and system based on pre-authorization supervision, aiming at improving payment security and user experience. The method first collects historical purchase behavior data and operation data of a target user, and constructs operation fingerprint features of the target user under different purchase behaviors. Through cluster analysis, the historical purchase operation diversity index of the target user in various purchase behaviors is calculated and determined to depict the behavior mode stability. Then, real-time purchase operation data of the target user is obtained, and a real-time diversity index is calculated and combined with historical data to evaluate the current purchase risk level of the target user. According to the risk result, pre-authorization supervision control is performed on the transaction between the user and the merchant account, so that a risk-controllable post-paid payment strategy is realized. The method can dynamically identify abnormal transaction behaviors, enhance the risk prevention ability of the payment system, and is suitable for various online and offline intelligent payment scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A flowchart of a post-paid intelligent payment method based on pre-authorization supervision is shown. Figure 2 A flowchart of real-time purchase risk of a target user is shown. Figure 3 A flowchart of constructing a post-paid strategy of a target user is shown. Figure 4 A block diagram of a post-paid intelligent payment system based on pre-authorization supervision is shown. DETAILED DESCRIPTION

[0015] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0016] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below.

[0017] Figure 1 A flowchart of a post-paid intelligent payment method based on pre-authorization supervision is shown.

[0018] As Figure 1 shown, the first aspect of the present application provides a post-paid intelligent payment method based on pre-authorization supervision, comprising: S102, obtaining historical purchase behavior data and historical purchase operation data of a target user, and constructing purchase operation fingerprints of the target user under different purchase behaviors; S104, the purchase operation fingerprint is clustered, and a historical purchase operation diversity index of different purchase behaviors of the target user is determined; S106, a real-time purchase operation diversity index of the target user is acquired, and a real-time purchase risk of the target user is determined according to the historical purchase operation diversity index and the real-time purchase operation diversity index; S108, the target user and the corresponding merchant account are pre-authorized and supervised according to the real-time purchase risk, and a post-payment strategy of the target user is constructed.

[0019] It should be noted that by acquiring the historical purchase behavior data and the historical purchase operation data of the target user, and constructing the purchase operation fingerprint of the user in different purchase scenarios, the personalized operation habits and behavior characteristics of the user can be effectively extracted, and the accuracy of user identity recognition is enhanced. Secondly, by clustering analysis on the purchase operation fingerprint, the historical behavior diversity index of the user in different purchase types is further refined, the real-time purchase operation diversity index is acquired in the transaction process, and dynamic comparison is made with the historical diversity index, so that whether the current operation of the user exists abnormal behavior can be identified in real time, so that the rapid early warning and fine discrimination of potential fraud behavior are realized. Finally, the fund flow between the user and the merchant is pre-authorized and supervised in combination with the real-time purchase risk value, by setting reasonable freezing proportion and supervision level, it is ensured that the fund is in a controlled state before the transaction is completed, and the fund loss caused by non-delivery of goods, service quality dispute or user fraud is avoided.

[0020] According to the embodiment of the application, the historical purchase behavior data and the historical purchase operation data of the target user are acquired, and the purchase operation fingerprint of different purchase behaviors of the target user is constructed, specifically: The historical purchase behavior data of the target user is acquired, and the historical purchase behavior data includes purchase commodity type, purchase price, purchase frequency data; The historical purchase operation data of the target user for different purchase behaviors is acquired, and the historical purchase operation data includes the product search and selection operation within a preset time period before the user completes the product purchase, the click and browsing operation of each information display board of the product, the customer service inquiry operation, the collection and purchase operation, the payment operation, and the operation time length information of each purchase operation type is acquired; Each purchase operation type of the historical purchase operation data is sorted according to the operation timestamp, the purchase operation sequence is constructed, and the feature extraction of the purchase operation sequence is performed, including the duration of each purchase operation type, the interval time between adjacent operations, the operation type conversion frequency, the operation repetition number, and the page scrolling speed; The feature vector of the purchase operation sequence feature is constructed, a hidden Markov model is introduced to model the feature vector, a purchase operation sequence corresponding to each purchase behavior is taken as an observation sequence, and a probability distribution of the observation sequence is calculated through a forward-backward algorithm; A hidden state sequence of the purchase operation sequence is decoded according to the probability distribution through a Viterbi algorithm, the hidden state sequence is fused with a corresponding feature vector, a principal component analysis dimension reduction processing is performed on the fused feature vector, and a dimension-reduced feature vector is obtained. Hash values of each dimension-reduced feature vector are constructed based on hash coding, the hash values are constructed into a hash fingerprint set, and purchase operation fingerprints of different purchase behaviors of a target user are obtained.

[0021] It should be noted that the purchase operation sequence is composed of multiple operations executed by a user in a time sequence during shopping, and the duration, interval time, conversion frequency and the like of these operations constitute a high-dimensional feature vector. The hidden Markov model assumes that an observation sequence (i.e., a purchase operation feature actually executed by a user) is generated by a series of hidden states that cannot be directly observed, and calculates the probability distribution of the observation sequence through a forward-backward algorithm, thereby capturing the potential state change rule behind the user behavior. The hidden state sequence reflects the psychological and decision-making state of a user in different purchase behavior stages, and these states, although not directly observable, play a key role in understanding the user's operation habits and intentions. The hidden state sequence is decoded through a Viterbi algorithm, the hidden information is fused with the original feature vector, and the description ability of the purchase behavior is further improved. The fused feature vector is subjected to principal component analysis dimension reduction processing to remove redundant information and highlight the main behavior features, and finally the dimension-reduced features are converted into fixed-length hash values through hash coding technology to form unique and efficient purchase operation fingerprints. The hidden state sequence refers to a potential stage that is automatically learned from user operation behavior by a hidden Markov model (HMM) and cannot be directly observed but can explain the purchase decision-making process thereof, for example, may include a "price comparison browsing stage", a "hesitation decision-making stage", an "impulsive purchase stage" or a "repeated verification stage", and the like, and each state corresponds to different behavior patterns and risk features. The goods herein include items, courses, financial products, services and the like. The adding is adding to a shopping cart. The purchase operation fingerprint can accurately identify the operation mode of a user in different purchase behaviors.

[0022] According to an embodiment of the present application, the purchase operation fingerprint is clustered to determine a historical purchase operation diversity index of a target user for different purchase behaviors, specifically: The purchase operation fingerprint, i.e., the hash fingerprint set, is subjected to normalization preprocessing, the Hamming distance between each purchase operation fingerprint after normalization is calculated, and a distance matrix between each purchase operation fingerprint is constructed according to the Hamming distance. The density clustering algorithm is introduced, the initial neighborhood search radius ε is calculated according to the standard deviation of all purchase operation fingerprints in each dimension and the total number of fingerprint points, and the initial minimum number of points MinPts in the neighborhood is estimated based on the distance distribution in the distance matrix; All purchase operation fingerprint points are traversed, an unvisited point is selected as a core point, and all fingerprint points contained in the ε neighborhood of the core point are found by the neighborhood search radius ε to form a temporary subset; If the number of fingerprint points contained in the temporary subset is greater than or equal to the minimum number of points MinPts in the neighborhood, a new cluster is created, the core point is added to the new cluster, and recursive neighborhood expansion is performed on each fingerprint point in the temporary subset to check whether the number of points in the ε neighborhood of each adjacent point in the temporary subset reaches MinPts, if so, the adjacent point is marked as a new core point and its neighborhood range is expanded, until no new adjacent core point is found, thereby forming a complete cluster; If the number of points in the ε neighborhood of the core point is less than MinPts, the core point is marked as a boundary point, after the traversal of all core points and the generation of clusters, all boundary points are visited, and the distance from each boundary point to the core points in each generated cluster is checked, if there is a core point with a distance less than or equal to the neighborhood search radius ε from the boundary point, and the core point has been classified into a cluster, the boundary point is merged into the cluster, and the purchase operation fingerprint clustering result is obtained; The total number K of clusters in the purchase operation fingerprint clustering result is obtained, the cluster center of each cluster is calculated, and the internal dispersion of each cluster is calculated according to the average distance between all purchase operation fingerprints in the cluster and the cluster center; The historical purchase diversity index of the purchase operation fingerprint in the cluster is determined according to the product of the total number K of clusters and the weighted sum of the internal dispersion of each cluster, and the purchase behavior of each cluster is determined, and the historical purchase operation diversity index of the target user for different purchase behaviors is obtained.

[0023] It should be noted that all the purchase operation fingerprint hash values are normalized to eliminate the dimensional differences between different dimensions, a distance matrix is constructed by calculating the Hamming distance, and the similarity between the fingerprints is quantified. Then, the neighborhood search radius epsilon and the minimum number of neighborhood points MinPts are automatically determined based on the distribution characteristics of the fingerprint points, so as to ensure that the clustering parameters are adaptive and robust. In the clustering process, unvisited fingerprint points are selected as core points, and the generation and expansion of clusters are carried out according to the density of the points in the neighborhood, so as to effectively identify high-density clusters and boundary points, and avoid the influence of noise points on the clustering results. After the traversal of all core points is completed, the boundary points are further reasonably classified into adjacent clusters, so as to ensure the integrity of the clustering. In the clustering result, each cluster represents a typical purchase operation mode, and the cluster center reflects the behavior characteristics of the mode. The internal dispersion of the cluster is obtained by calculating the average distance between all fingerprints in the cluster and the cluster center, and the total number K of clusters is combined to form a weighted comprehensive historical purchase operation diversity index. The index quantifies the operation variability and behavior complexity of the user in different purchase behaviors, and can reflect the diversity and stability of the user behavior. The historical purchase operation diversity index of the target user for different purchase behaviors is the operation behavior diversity index of the user before purchasing different prices and different types and different purchase frequencies of goods. The purchase operation fingerprint clustering result classifies the operation behaviors of the user before purchasing different prices and different types and different purchase frequencies of goods into a category, and the clustering result shows what operation behavior habits the user usually has before purchasing different prices and different types and different purchase frequencies of goods. For example, when a target user purchases high-value electronic products, the purchase operation fingerprint clustering result of the user shows that the user usually performs multiple detailed product information browsing, compares different brands and models, frequently checks product reviews, and consults customer service multiple times, which shows a high degree of detail and caution; when the user purchases daily low-price household goods, the operation fingerprint of the user shows rapid browsing, a small amount of clicking, and a short dwell time, which shows obvious operation simplification and decision-making speed acceleration; in addition, for clothing goods in the medium price range, the user may show behavior habits such as collecting, adding to shopping cart, and occasionally querying for preferential activities. Through these clustering results, the typical operation habits of the user in different price ranges, different types of goods, and different purchase frequencies can be accurately described.

[0024] Figure 2 A flowchart of the real-time purchase risk of the target user of the application is shown.

[0025] According to an embodiment of the application, the real-time purchase operation diversity index of the target user is obtained, and the real-time purchase risk of the target user is determined according to the historical purchase operation diversity index and the real-time purchase operation diversity index, specifically: S202, acquire real-time purchase behavior data of the target user before confirming a commodity order and before payment of the commodity, and real-time purchase operation data of a preset time period before payment of the commodity; S204, construct a real-time purchase operation sequence according to the real-time purchase operation data, perform feature extraction on the real-time purchase operation sequence, and construct a real-time purchase operation feature vector; S206, perform similarity comparison between the real-time purchase operation feature vector and the feature vector after dimension reduction, determine the attribution probability distribution of the current real-time purchase operation feature vector in each cluster by calculating the distance between the real-time feature vector and the cluster center of each purchase operation fingerprint; S208, perform weighted fusion of the attribution probability distribution and the historical purchase operation diversity index corresponding to each cluster to construct a real-time purchase operation diversity index of the current purchase operation; S210, calculate the deviation degree between the real-time purchase operation diversity index and the historical purchase operation diversity index, and determine the real-time purchase risk of the target user according to the deviation degree.

[0026] It should be noted that by acquiring real-time purchase behavior data of the target user before confirming a commodity order and before payment of the commodity, constructing a real-time purchase operation sequence and extracting features thereof, forming a real-time purchase operation feature vector, and accurately depicting the current purchase behavior of the user. By performing similarity comparison between the real-time feature vector and the feature vector after dimension reduction of the historical purchase behavior, calculating the attribution probability distribution of the real-time behavior in each purchase operation fingerprint clustering cluster, the matching degree of the current purchase behavior of the user and the historical behavior mode can be effectively reflected. Based on the weighted fusion of the attribution probability distribution and the corresponding historical purchase operation diversity index, the real-time purchase operation diversity index is constructed to accurately quantify the diversity characteristics of the current behavior. Further, by calculating the deviation degree between the real-time purchase operation diversity index and the historical diversity index, the abnormal change of the purchase behavior of the user can be identified in time, and then the real-time purchase risk can be accurately evaluated.

[0027] Figure 3 A flowchart for constructing a post-payment strategy of a target user is shown.

[0028] According to the embodiment of the present application, the target user and the corresponding merchant account are pre-authorized and supervised according to the real-time purchase risk, and the post-payment strategy of the target user is constructed, specifically: According to the real-time purchase risk, if the real-time purchase risk is greater than a preset risk value, a post-payment pre-authorization instruction is initiated to the user payment account, and when the post-payment pre-authorization instruction of the user is received, the supervision level of the user payment account is determined according to the real-time purchase risk; According to the supervision level, the commodity payment amount freezing ratio and freezing duration of the user account are determined to obtain post-payment pre-authorization supervision parameters; According to the post-paid pre-authorization supervision parameter, the user account is pre-authorized, the smart contract is initialized according to the post-paid pre-authorization supervision parameter, the smart contract includes user identity, merchant account information, commodity amount details, commodity description and pre-authorization supervision parameter, the smart contract is bound with the preset account; According to the smart contract, the pre-authorization operation is executed, the commodity payment amount of the user is transferred from the user account to the preset account according to the freezing ratio to freeze, a unique pre-authorization voucher is generated, and the freezing state and freezing time are recorded, and the smart contract, the unique pre-authorization voucher, the freezing state and the freezing time are synchronized to the preset insurance account; Real-time monitoring of merchant delivery and user receipt state data, the merchant delivery and user receipt state data includes logistics state, merchant delivery voucher, user signing state; According to the merchant delivery and user receipt state data, the user's commodity delivery is determined, if the commodity has been delivered to the user, the payment instruction is executed based on the smart contract and the supervision level, the frozen fund is transferred to the merchant account, if the payment is abnormal, the frozen amount is transferred to the preset insurance account, and the merchant is compensated according to the preset insurance account; If the user does not confirm receipt or raise objection within the preset time, the platform is responsible for the judgment, the delivery is carried out according to the judgment result, and the post-paid strategy of the target user is obtained.

[0029] It should be noted that the preset account includes a central bank digital RMB MO special account; by implementing pre-authorization supervision on the target user and the corresponding merchant account according to the real-time purchase risk, a dynamic and intelligent post-payment strategy is constructed, and effective control of transaction risk and fund security protection are realized. When the real-time purchase risk exceeds the preset threshold, the system automatically initiates a post-payment pre-authorization instruction to the user's payment account, and determines the corresponding supervision level based on the risk level, so as to flexibly set the freezing ratio and freezing time of the commodity payment amount. Through the smart contract technology, the user identity, merchant account information, commodity details and pre-authorization supervision parameters are bound and executed, the automation and transparency of fund freezing are realized, a unique pre-authorization voucher is generated and synchronized to the preset insurance account, and the credibility and security of fund supervision are enhanced. The system monitors the commodity delivery and user receipt state in real time, ensures that only after confirming that the commodity is successfully delivered to the user, the payment instruction is automatically executed to transfer the fund to the merchant account, effectively preventing the fund risk caused by non-delivery or disputes. If the payment is abnormal, the frozen fund is timely transferred to the insurance account, and the insurance mechanism is used for bottom-up compensation to protect the interests of the merchant. If the user does not confirm the receipt in time or raise objections, the platform reasonably handles according to the responsibility determination result, and further improves the transaction closed loop. At the risk control level, the system dynamically adjusts the supervision level according to the real-time risk assessment result, and differentiates the fund freezing ratio and freezing time, which effectively prevents high-risk transactions and avoids excessive intervention on normal transactions; at the fund security level, the central bank digital RMB MO special account is used for fund freezing, which ensures that the pre-authorization fund is isolated from the user's main account and is special and special, and the automatic execution of fund transfer rules through the smart contract eliminates the risk of human operation; at the transaction guarantee level, an insurance bottom-up mechanism and platform arbitration rules are established, and when there is a delivery dispute, the pre-set insurance account can be quickly compensated, and the responsibility can be divided according to the platform determination result. The said according to the determination result to deliver specifically is: if the payment is abnormal, the smart contract will start the insurance bottom-up compensation to the merchant. In the transaction, if one party has objections to the transaction, leading to the inability of the transaction parties to reach a consensus, the platform determines the responsibility. If the responsibility is on the user side and the user has no reason to refuse to pay, the insurance bottom-up process is introduced to pay the merchant. The insurance will be responsible for the user. If the responsibility is on the merchant side, the system automatically unfreezes the user fund, and the insurance will be responsible for the merchant.

[0030] According to the embodiment of the application, the pre-authorization supervision on the target user and the corresponding merchant account according to the real-time purchase risk further comprises: When the real-time purchase risk of the preset number of user accounts continuously exceeds the preset risk value, the historical pre-authorization failure records and corresponding risk fluctuation data of the user account are obtained, and the deviation coefficient of the historical risk fluctuation variance and the current real-time purchase risk is calculated. If the deviation coefficient exceeds the dynamic threshold, an interaction request containing a behavior verification question is sent to the user terminal, the operation response sequence of the user to the verification question is obtained, and the operation interval time, input correction times and trajectory features in the response sequence are extracted; When the operation interval time is less than the historical average response time and the input correction times are lower than the abnormal threshold, it is determined that the operation is artificial, the pre-authorization instruction is continued to be executed, otherwise the biometric feature verification is started, the real-time face image and voiceprint feature of the user are obtained, and the real-time face image and voiceprint feature are matched with the pre-stored template; When the matching similarity is lower than the first threshold but higher than the second threshold, the location of the payment account associated device is verified, and if the deviation distance of the device location from the commonly used geofence exceeds the preset range, the account is frozen and a security alert is sent; The real-time purchase risk value is re-evaluated according to the operation response sequence, biometric feature verification and location verification, and if the risk value decreases by more than the drawdown threshold, the freezing proportion is reduced and the freezing duration is shortened, otherwise the original pre-authorization supervision parameters are maintained.

[0031] It should be noted that during the user purchase process, the real-time purchase risk of the preset number of purchase operations may continuously exceed the preset risk value. By analyzing the historical pre-authorization failure records and risk fluctuations of the user account, the system can calculate the deviation degree of the risk and timely identify abnormal risk fluctuations. Once the deviation exceeds the dynamic threshold, the system actively sends a behavior verification question to the user terminal, collects the operation response sequence of the user, and judges whether the operation is a real human operation through behavior features such as operation interval time and input correction times, further reducing the risk of automated attacks. For abnormal operations, the system starts a higher level of biometric feature verification, compares the real-time collected face image and voiceprint information with the pre-stored template, combines with the device location verification, comprehensively judges the authenticity of the user's identity and the safety of the operation environment, and prevents identity theft and abnormal use of the account. Based on the multiple verification results, the system dynamically adjusts the real-time purchase risk evaluation, and flexibly adjusts the freezing proportion and duration of the account according to the risk change, achieving a balance between risk control and user experience. This method effectively enhances the account security protection capability, improves the accuracy and response speed of risk identification, and reduces the misjudgment rate. The pre-stored template refers to the user biometric feature data collected and saved by the system in advance, which is used for identity verification by comparing with the real-time collected biometric features.

[0032] According to the embodiments of the present application, the method further comprises: A real-time feature vector of the latest purchase operation sequence of the target user is obtained, the belonging probability distribution of the real-time vector in each cluster is calculated according to the distance between the real-time feature vector and the cluster center of the historical purchase operation fingerprint, and a behavior drift factor is constructed based on the dimension compression rate of the real-time operation sequence and the change amount of the hidden Markov state transition probability matrix; According to the deviation degree of the home probability distribution and the historical purchase operation diversity index and the behavior drift factor, a historical reference weight decay coefficient is calculated, and when the decay coefficient is lower than a dynamic threshold, an operation sequence set of the user in a recent preset time period is acquired; From the recent operation sequence set, a median of operation interval time, a variance of page stay duration, and a state jump frequency are extracted to construct a short-term behavior feature matrix. According to the KL divergence of the short-term behavior feature matrix and the historical purchase operation fingerprint, a behavior offset direction angle is determined, the original clustering cluster center is rotated and transformed based on the behavior offset direction angle, and an adaptive cluster center coordinate is generated. The real-time feature vector and the adaptive cluster center coordinate are recalculated to obtain a home probability distribution, and a calibrated real-time purchase operation diversity index is output according to the modified home probability distribution and the historical purchase operation diversity index.

[0033] According to the embodiment of the present application, the historical reference weight decay coefficient is calculated, specifically: According to the dimension compression rate in the behavior drift factor, a shrinkage ratio of the real-time operation feature vector relative to the historical feature vector dimension is calculated, and an operation logic variation degree is obtained through the change amount of the hidden Markov state transition probability matrix; The drift intensity coefficient is generated by fusing the shrinkage ratio and the operation logic variation degree, and the decay calculation is triggered when the drift intensity coefficient exceeds a first critical value; The time window weight base is determined according to the ratio of the number of operation sequences in the recent preset time period to the total amount of historical operation sequences, and the time decay correction amount is calculated by combining the drift intensity coefficient and the standard deviation of the historical purchase operation diversity index; The time decay coefficient is obtained by dynamically decaying the time weight of the original historical data based on the time decay correction amount, and the time decay coefficient is applied to the historical purchase operation diversity index; The historical purchase operation diversity index is weighted and smoothed according to the time decay coefficient to generate an adaptive historical diversity baseline; The real-time purchase risk stage is determined, the relative entropy of the real-time purchase operation diversity index and the adaptive historical diversity baseline is taken as a deviation degree measurement index, the risk judgment boundary is adjusted in combination with the projection length of the behavior offset direction angle on the feature space, and when the projection length exceeds a second critical value, a negative compensation factor is applied to the fund freezing ratio in the pre-authorization supervision parameter.

[0034] It should be noted that since the historical purchase operation fingerprint is constructed based on early behavior data of the user, when the user has a persistent behavior change due to a change in life state or a change in consumption habits, the system will still compare the real-time operation features with the outdated historical pattern. This static matching mechanism will mistakenly amplify the behavior difference, misjudge the normal consumption habits as high-risk operations. This not only triggers unnecessary pre-authorization supervision measures, increases the misjudgment rate of fund freezing, but also reduces the user experience and trust of the payment system. By introducing the construction of the behavior drift factor, the distance between the real-time feature vector and the historical cluster center is attributed to the probability distribution, and the dimension compression rate and the change of the hidden Markov state transition probability matrix are used to quantify the dynamic drift of the user behavior. The change trend of the user behavior pattern is effectively captured. Further, by extracting the short-term behavior features of the recent operation sequence set and analyzing the difference with the historical purchase operation fingerprint, the cluster center position is dynamically adjusted to realize the generation of adaptive cluster center, and the accuracy of the attribution probability distribution is improved. Based on this, the historical reference weight decay coefficient is used to dynamically time-weight the historical purchase operation diversity index, which can smooth the influence of historical data on risk assessment and enhance the sensitivity and adaptability of the system to user behavior drift. In addition, combined with the dynamic adjustment of the risk judgment boundary by the behavior offset direction angle, more accurate deviation measurement and risk stage division in the feature space are realized, so as to reasonably adjust the fund freezing ratio in the pre-authorization supervision parameter. The risk identification accuracy and the flexibility of pre-authorization supervision of the post-paid intelligent payment method when facing user behavior changes are significantly improved, and the misjudgment and misrefusal risks are reduced.

[0035] Figure 4 A block diagram of a post-paid intelligent payment system based on pre-authorization supervision is shown.

[0036] The second aspect of the present application also provides a post-paid intelligent payment system 4 based on pre-authorization supervision, which comprises a memory 41 and a processor 42, wherein the memory comprises a post-paid intelligent payment method program based on pre-authorization supervision, and the post-paid intelligent payment method program based on pre-authorization supervision is executed by the processor to realize the following steps: Obtain the historical purchase behavior data and the historical purchase operation data of the target user, and construct the purchase operation fingerprint of different purchase behaviors of the target user; Cluster the purchase operation fingerprint to determine the historical purchase operation diversity index of the target user for different purchase behaviors; Obtain the real-time purchase operation diversity index of the target user, and determine the real-time purchase risk of the target user according to the historical purchase operation diversity index and the real-time purchase operation diversity index; According to the real-time purchase risk, pre-authorization supervision is performed on the target user and the corresponding merchant account, and a post-paid strategy of the target user is constructed.

[0037] The application discloses a postpaid intelligent payment method and system based on pre-authorization supervision, aiming to improve payment security and user experience. The method first collects historical purchase behavior data and operation data of a target user, and constructs operation fingerprint features of the user under different purchase behaviors. Through cluster analysis, the historical purchase operation diversity index of the user in various purchase behaviors is calculated and determined to depict the behavior mode stability. Then, real-time purchase operation data of the user is obtained, the real-time diversity index is calculated, and the current purchase risk level of the user is evaluated in combination with the historical data. According to the risk result, the transaction between the user and the merchant account is pre-authorized and supervised, so that the postpaid payment strategy with controllable risk is realized. The method can dynamically identify abnormal transaction behaviors, enhance the risk prevention ability of the payment system, and is suitable for various online and offline intelligent payment scenarios.

[0038] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative, for example, the division of the units is only a logical function division, and actual implementation can have another division mode, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling or communication connection between each component part shown or discussed can be through some interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0039] The units described above as separate components can or can not be physically separated, and the components shown as units can or can not be physical units; they can be located in one place or distributed on multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0040] In addition, each functional unit in each embodiment of the application can be integrated in one processing unit, or each unit can be a separate unit, or two or more units can be integrated in one unit; the integrated unit can be realized in the form of hardware or hardware plus software functional unit.

[0041] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware, the foregoing program can be stored in a computer readable storage medium, and the program executes the steps of the method embodiments when executed; and the foregoing storage medium includes a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disc or an optical disc, and various storage medium capable of storing program codes.

[0042] Alternatively, the integrated unit of the present application can be stored in a computer readable storage medium if it is realized in the form of a software function module and sold or used as an independent product. Based on such understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the embodiments of the present application. The foregoing storage medium includes a mobile storage device, a ROM, a RAM, a magnetic disc or an optical disc, and various storage medium capable of storing program codes.

[0043] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A post-paid intelligent payment method based on pre-authorization supervision, characterized in that, The method comprises the following steps: obtaining historical purchase behavior data and historical purchase operation data of a target user, and constructing purchase operation fingerprints of different purchase behaviors of the target user; performing clustering operation on the purchase operation fingerprints to determine a historical purchase operation diversity index of the target user for different purchase behaviors; obtaining a real-time purchase operation diversity index of the target user, and determining a real-time purchase risk of the target user according to the historical purchase operation diversity index and the real-time purchase operation diversity index; performing pre-authorization supervision on the target user and a corresponding merchant account according to the real-time purchase risk, and constructing a postpaid strategy of the target user.

2. The postpaid intelligent payment method based on pre-authorization supervision according to claim 1, characterized in that, The step of obtaining historical purchase behavior data and historical purchase operation data of a target user, and constructing purchase operation fingerprints of different purchase behaviors of the target user comprises the following steps: obtaining historical purchase behavior data of the target user, wherein the historical purchase behavior data comprises purchase product type, purchase price and purchase frequency data; obtaining historical purchase operation data of the target user for different purchase behaviors, wherein the historical purchase operation data comprises product search and selection operation, product information display board clicking and browsing operation, customer service inquiry operation, collection and purchase adding operation, and payment operation within a preset time period before the user completes product purchase, and operation time length information of each purchase operation type is obtained; sorting each purchase operation type of the historical purchase operation data according to operation timestamp, constructing a purchase operation sequence, and performing feature extraction on the purchase operation sequence, including duration of each purchase operation type, interval time between adjacent operations, operation type conversion frequency, operation repetition number, and page scrolling speed; constructing a feature vector from the purchase operation sequence features, introducing a hidden Markov model to model the feature vector, taking the purchase operation sequence corresponding to each purchase behavior as an observation sequence, and calculating probability distribution of the observation sequence through a forward-backward algorithm; decoding the hidden state sequence of the purchase operation sequence through a Viterbi algorithm according to the probability distribution, fusing the hidden state sequence and the corresponding feature vector, and performing principal component analysis dimension reduction processing on the fused feature vector to obtain a dimension-reduced feature vector; constructing a hash value of each dimension-reduced feature vector based on hash coding, constructing a hash fingerprint set from the hash values, and obtaining purchase operation fingerprints of different purchase behaviors of the target user.

3. The postpaid intelligent payment method based on pre-authorization supervision according to claim 1, characterized in that, The step of performing clustering operation on the purchase operation fingerprints to determine a historical purchase operation diversity index of the target user for different purchase behaviors comprises the following steps: performing normalization preprocessing on the purchase operation fingerprints, i.e. the hash fingerprint set, calculating Hamming distance between each purchase operation fingerprint after normalization, and constructing a distance matrix between each purchase operation fingerprint according to the Hamming distance; introducing a density clustering algorithm, calculating an initial neighborhood search radius ε according to standard deviation of all purchase operation fingerprints in each dimension and total number of fingerprint points, and estimating the initial minimum number of points MinPts in the neighborhood based on the distance distribution percentile in the distance matrix; Traverse all purchase operation fingerprint points, select a point not yet visited as a core point, find all fingerprint points contained in the core point ε neighborhood through neighborhood search radius ε to form a temporary subset; If the number of fingerprint points contained in the temporary subset is greater than or equal to the minimum number of points MinPts in the neighborhood, a new cluster is created, the core point is added to the new cluster, and the neighborhood of each fingerprint point in the temporary subset is recursively expanded to check whether the number of points in the ε neighborhood of each adjacent point in the temporary subset reaches MinPts, if so, continue to mark the adjacent point as a new core point and expand its neighborhood range until no new adjacent core point is found, thereby forming a complete cluster; If the number of points in the core point ε neighborhood is less than MinPts, the core point is marked as a boundary point, after the traversal of all core points and the generation of clusters, all boundary points are accessed, and the distance from each boundary point to the core points in each generated cluster is checked in turn, if there is a core point with a distance less than or equal to the neighborhood search radius ε from the boundary point, and the core point has been classified into a cluster, the boundary point is merged into the cluster, and the purchase operation fingerprint clustering result is obtained; Obtain the total number K of clusters in the purchase operation fingerprint clustering result, calculate the cluster center of each cluster, and calculate the internal dispersion of each cluster according to the average distance between all purchase operation fingerprints in the cluster and the cluster center; According to the product of the total number K of clusters and the weighted sum of the internal dispersion of each cluster, the historical purchase diversity index of the purchase operation fingerprint in the cluster is determined, and the purchase behavior of each cluster is determined to obtain the historical purchase operation diversity index of the target user for different purchase behaviors.

4. The postpaid intelligent payment method based on pre-authorization supervision according to claim 2, characterized in that, The real-time purchase risk of the target user is determined according to the historical purchase operation diversity index and the real-time purchase operation diversity index, specifically: Obtain the real-time purchase behavior data of the target user before confirming the commodity order and the real-time purchase operation data of the target user in a preset time period before the commodity payment; According to the real-time purchase operation data, a real-time purchase operation sequence is constructed, feature extraction is performed on the real-time purchase operation sequence, and a real-time purchase operation feature vector is constructed; The real-time purchase operation feature vector and the reduced feature vector are compared in similarity, the distance between the real-time feature vector and the cluster center of each purchase operation fingerprint is calculated to determine the belonging probability distribution of the current real-time purchase operation feature vector in each cluster; According to the belonging probability distribution and the historical purchase operation diversity index corresponding to each cluster, a real-time purchase operation diversity index of the current purchase operation is constructed by weighted fusion; The deviation degree between the real-time purchase operation diversity index and the historical purchase operation diversity index is calculated, and the real-time purchase risk of the target user is determined according to the deviation degree.

5. The pre-authorization supervision based postpaid intelligent payment method according to claim 1, characterized in that, According to the real-time purchase risk, the target user and the corresponding merchant account are pre-authorized and supervised, and a post-payment strategy of the target user is constructed, specifically: According to the real-time purchase risk, if the real-time purchase risk is greater than a preset risk value, a post-payment pre-authorization instruction is initiated to a user payment account, and when the post-payment pre-authorization instruction of the user is received, a supervision level of the user payment account is determined according to the real-time purchase risk; According to the supervision level, a commodity payment amount freezing ratio and a freezing duration of the user account are determined to obtain post-payment pre-authorization supervision parameters; According to the post-payment pre-authorization supervision parameters, the user account is pre-authorized and supervised, and an intelligent contract is initialized according to the post-payment pre-authorization supervision parameters, the intelligent contract includes a user identity, merchant account information, commodity amount details, commodity description, and pre-authorization supervision parameters, and the intelligent contract is bound to a preset account; According to the intelligent contract, a pre-authorization operation is performed, the commodity payment amount of the user is transferred from the user account to the preset account according to the freezing ratio to freeze, a unique pre-authorization voucher is generated, and the freezing state and freezing time are recorded, and the intelligent contract, the unique pre-authorization voucher, the freezing state and the freezing time are synchronized to a preset insurance account; Real-time monitoring of merchant delivery and user receipt state data, the merchant delivery and user receipt state data includes logistics state, merchant delivery voucher, and user signature state; According to the merchant delivery and user receipt state data, the commodity delivery of the user is determined, if the commodity has been delivered to the user, the payment instruction is executed based on the intelligent contract and the supervision level, the frozen fund is transferred to the merchant account, if the payment is abnormal, the frozen amount is transferred to the preset insurance account, and the merchant is compensated according to the preset insurance account; If the user does not confirm receipt or raise objection within a preset time, the responsibility is determined according to the platform, and the delivery is performed according to the determination result to obtain the post-payment strategy of the target user.

6. The pre-authorization supervision based postpaid intelligent payment method according to claim 5, characterized in that, The pre-authorization supervision of the target user and the corresponding merchant account according to the real-time purchase risk to build the post-payment strategy of the target user also includes: When the real-time purchase risk of the preset number of user accounts exceeds the preset risk value, the historical pre-authorization failure record and the corresponding risk fluctuation data of the user account are obtained, the deviation coefficient of the historical risk fluctuation variance and the current real-time purchase risk is calculated; If the deviation coefficient exceeds a dynamic threshold, an interaction request containing a behavior verification question is sent to the user terminal, the operation response sequence of the user to the verification question is obtained, and the operation interval time, input correction times and trajectory features in the response sequence are extracted; When the operation interval time is less than the historical average response time and the input correction times are lower than the abnormal threshold, it is determined as a manual operation, and the pre-authorization instruction is continued to be executed, otherwise the biological feature verification is started, and the real-time face image and voiceprint feature of the user are obtained; The real-time biological features are matched with the pre-stored templates, when the matching similarity is lower than a first threshold but higher than a second threshold, the location of the payment account associated device is verified, if the device location deviates from the common geographic fence by more than a preset range, the account is frozen and a security alarm is sent. According to the operation response sequence, the biometric verification, and the location verification, the real-time purchase risk value is re-evaluated, if the risk value decreases by more than a retreat threshold, the freezing proportion is reduced and the freezing duration is shortened, otherwise the original pre-authorization supervision parameters are maintained.

7. A post-paid intelligent payment system based on pre-authorization supervision, characterized in that, The postpaid intelligent payment system based on pre-authorization supervision comprises a storage and a processor, the storage comprises a postpaid intelligent payment method program based on pre-authorization supervision, and the postpaid intelligent payment method program based on pre-authorization supervision is executed by the processor to implement the following steps: Obtain historical purchase behavior data and historical purchase operation data of a target user, and construct purchase operation fingerprints of different purchase behaviors of the target user; Cluster the purchase operation fingerprints to determine a historical purchase operation diversity index of the target user for different purchase behaviors; Obtain a real-time purchase operation diversity index of the target user, and determine a real-time purchase risk of the target user according to the historical purchase operation diversity index and the real-time purchase operation diversity index; According to the real-time purchase risk, pre-authorization supervision is performed on the target user and a corresponding merchant account, and a postpaid strategy of the target user is constructed.

Images