Method, device, electronic device and storage medium for identifying merchant's behavior of skipping payment

Abstract

The present application discloses a method for identifying a merchant's behavior of running away from a bill, which belongs to the field of computer technology and is helpful to improve the accuracy of identifying a merchant's behavior of running away from a bill. The method for identifying a merchant's behavior of running away from a bill disclosed in the present application includes: determining the description features of the running away behavior of at least one item of evidence of the target merchant's behavior of running away from a bill based on an expert model; predicting the probability of running away from a bill corresponding to each item of evidence of the target merchant's behavior of running away from a bill based on the expert model and the description features of the at least one item of evidence of the target merchant's behavior of running away from a bill; determining the first probability of running away from a bill of the target merchant by linearly fusing the probability of running away from a bill of each item of evidence of the target merchant's behavior of running away from a bill; predicting the probability of running away from a bill of the target merchant based on multiple preset machine learning models, and obtaining the second probability of running away from a bill corresponding to each of the machine learning models; calculating the probability of running away from a bill of the target merchant through a preset fusion model according to the first probability of running away from a bill and the second probability of running away from a bill.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, electronic device and computer-readable storage medium for identifying a merchant's behavior of skipping a payment. Background Art

[0002] E-commerce platforms should provide convenience and influence for merchants on the platform, and achieve user flow and import for merchants on the platform. The platform and merchants settle resources based on resource settlement data such as order volume completed through the platform. However, merchants' evasion of orders affects resource settlement data such as order volume through various behaviors, thereby affecting the accuracy of resource settlement between the platform and merchants, which has a great impact on the platform's revenue.

[0003] To this end, methods for identifying merchants who have escaped bills have emerged in the prior art, mainly including: manually determining merchants who have escaped bills based on user feedback; identifying merchants who have escaped bills based on expert models; predicting the probability of merchants who have escaped bills through machine learning models, and identifying merchants who have escaped bills based on the probability of escaped bills. However, the methods for identifying merchants who have escaped bills in the prior art have at least the following defects: first, the accuracy of merchants who have escaped bills is manually determined based on user feedback, and merchants who have escaped bills cannot be actively discovered; second, the identification of merchants who have escaped bills based on expert models does not have the ability to learn and cannot identify the endless number of behaviors of escaping bills; third, the method of predicting the probability of merchants who have escaped bills through machine learning models has a cold start problem, and when the number of samples is small, the recognition accuracy is low.

[0004] It can be seen that the existing method for identifying merchant's behavior of running away from the bill has at least the defect of low recognition accuracy. Summary of the invention

[0005] The present application provides a method for identifying merchant's behavior of running away from the bill, which helps to improve the accuracy of identifying merchant's behavior of running away from the bill.

[0006] In order to solve the above problems, in a first aspect, an embodiment of the present application provides a method for identifying a merchant's behavior of escaping from paying, including:

[0007] Determine, based on the expert model, a descriptive feature of at least one evidence item of the target merchant's behavior of running away from the bill;

[0008] Based on the expert model and the description features of the at least one item of evidence of the escape behavior, respectively predict the escape probability of the target merchant corresponding to each item of evidence of the escape behavior;

[0009] Determine a first probability of skipping a bill for the target merchant by linearly fusing the skipping bill probabilities corresponding to each of the skipping bill behavior evidence items of the target merchant;

[0010] Based on a plurality of preset machine learning models, respectively predict the probability of skipping the order for the target merchant, and obtain a second skipping probability corresponding to each of the machine learning models;

[0011] According to the first probability of escaping the bill and the second probability of escaping the bill corresponding to each of the machine learning models, the probability of escaping the bill of the target merchant is calculated through a preset fusion model.

[0012] In a second aspect, an embodiment of the present application provides a device for identifying a merchant's behavior of escaping from paying a bill, including:

[0013] A module for determining descriptive features of a behavior of skipping a bill, used to determine descriptive features of a behavior of skipping a bill of at least one evidence item of a behavior of skipping a bill of a target merchant based on an expert model;

[0014] a single evidence item escape probability determination module, configured to predict the escape probability of the target merchant corresponding to each escape behavior evidence item based on the expert model and the escape behavior description feature of the at least one escape behavior evidence item;

[0015] A first skipping probability determination module, configured to determine a first skipping probability of the target merchant by linearly fusing the skipping probabilities corresponding to the skipping behavior evidence items of the target merchant;

[0016] A second escape probability determination module, configured to predict the escape probability of the target merchant based on a plurality of preset machine learning models, and obtain second escape probabilities corresponding to the machine learning models;

[0017] The escape probability fusion module is used to calculate the escape probability of the target merchant through a preset fusion model based on the first escape probability and the second escape probability corresponding to the machine learning model.

[0018] On the third aspect, an embodiment of the present application further discloses an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the method for identifying the merchant's behavior of escaping the payment as described in the embodiment of the present application is implemented.

[0019] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the method for identifying merchant's behavior of running away from the bill disclosed in an embodiment of the present application.

[0020] The embodiment of the present application discloses a method for identifying a merchant's escape behavior, which comprises determining the escape behavior description features of at least one escape behavior evidence item of a target merchant based on an expert model; predicting the escape probability of the target merchant corresponding to each escape behavior evidence item based on the expert model and the escape behavior description features of the at least one escape behavior evidence item; determining a first escape probability of the target merchant by linearly fusing the escape probabilities corresponding to each escape behavior evidence item of the target merchant; predicting the escape probability of the target merchant based on a plurality of preset machine learning models to obtain second escape probabilities corresponding to each machine learning model; and calculating the escape probability of the target merchant through a preset fusion model based on the first escape probability and the second escape probability corresponding to each machine learning model, thereby solving the problem of at least low recognition accuracy in the prior art method for identifying a merchant's escape behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0022] Figure 1 This is a flow chart of a method for identifying a merchant's behavior of skipping a payment in Embodiment 1 of the present application;

[0023] Figure 2 This is a flow chart of a method for identifying a merchant's behavior of skipping a payment in Embodiment 2 of the present application;

[0024] Figure 3 This is one of the structural schematic diagrams of the device for identifying the merchant's behavior of skipping a payment in the third embodiment of the present application;

[0025] Figure 4 This is the second structural schematic diagram of the device for identifying the merchant's behavior of running away from the bill in the third embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0027] Embodiment 1

[0028] This embodiment discloses a method for identifying a merchant's behavior of skipping a payment. Figure 1As shown, the method includes: step 110 to step 150.

[0029] Step 110: Determine, based on the expert model, descriptive features of at least one item of evidence of a runaway behavior of the target merchant.

[0030] The item of evidence of the act of skipping a bill described in the embodiment of the present application is a description of identifying the target merchant as a merchant who skips a bill, and the item of evidence of the act of skipping a bill includes a description feature of the act of skipping a bill. The description feature of the act of skipping a bill is data support for identifying the target merchant as a merchant who skips a bill, for example, the probability of a user visiting both online and offline channels and not purchasing, the probability of a refund, the number of refunded orders, etc.

[0031] The expert model described in the embodiments of the present application may be a data processing model established based on expert experience data, or may be an expert experience value, etc. For example, the expert model includes: a feature processing template, an evidence item generation template, a number of preset features that can be used as features for describing behaviors of fleeing the bill, a method for determining features for describing behaviors of fleeing the bill, a method for setting feature thresholds, etc. The expert model also includes at least one feature combination for describing behaviors of fleeing the bill for determining merchants who flee the bill, each of the feature combination for describing behaviors of fleeing the bill corresponds to an evidence item for determining the merchant's fleeing the bill, and each of the feature combination for describing behaviors of fleeing the bill includes a main feature (i.e., the first feature for fleeing the bill referred to below) and multiple auxiliary features (i.e., the features for describing behaviors of fleeing the bill associated with the first feature for fleeing the bill referred to below).

[0032] Among them, the first bill-escape feature is a bill-escape behavior description feature (see below) determined based on the analysis results of the performance of the bill-escape merchant and the non-bill-escape merchant on several preset features, and the auxiliary feature is a bill-escape behavior description feature associated with the first bill-escape feature in the corresponding combination. In other embodiments of the present disclosure, the main feature and the auxiliary feature can also be determined based on expert experience.

[0033] In some embodiments of the present application, before the step of determining the descriptive features of at least one item of evidence of escape from the bill of the target merchant based on the expert model, the step further includes: obtaining preset features of several merchants who escape from the bill and the preset features of several merchants who are not merchants who escape from the bill; determining, based on the analysis results of the performance of the merchants who escape from the bill and the merchants who are not merchants who escape from the bill on each of the preset features, a preset number of the preset features with relatively large performance differences between the merchants who escape from the bill and the merchants who are not merchants who escape from the bill as first features of escape from the bill; for each of the first features of escape from the bill, determining the preset features associated with the first feature of escape from the bill, and using the preset features associated with the first feature of escape from the bill and the first feature of escape from the bill as descriptive features of the item of evidence of escape from the bill corresponding to the first feature of escape from the bill; for each of the first features of escape from the bill, determining the feature value interval and preset feature threshold of each feature of the feature of the item of evidence of escape from the bill corresponding to the first feature of escape from the bill; for each item of evidence of escape from the bill, constructing a feature processing template matching the item of evidence of escape from the bill according to the feature value interval and expert knowledge of each feature of the feature of the item of evidence of escape from the bill.

[0034] First, preset features are determined based on expert experience, and the preset features are features whose values ​​may change due to the merchant's behavior of escaping the order. The preset features include: merchant features, user features, and order transaction features.

[0035] In some embodiments of the present application, the user characteristics further include: user feedback characteristics, user online behavior characteristics, and user access characteristics. Among them, user feedback characteristics include whether the user has visited the store offline, whether the user's arrival time is a legal reception time, etc.; user online behavior characteristics include user purchase frequency, whether the user has a refund, whether the user is suspected of escaping the order, and whether the user is suspected of malicious claims; user access characteristics include monthly store visits, user online visit frequency, user offline visit frequency, and online and offline visit frequency without purchase, etc.

[0036] In some embodiments of the present application, merchant characteristics further include: store characteristics, surrounding competitor characteristics, store offline transaction characteristics, competitor cooperation characteristics, project characteristics, etc. Among them, store characteristics include whether it is a new store, whether there is wifi, whether there is a parking space, store average price, total number of store reviews, store praise rate, store rating, etc.; surrounding competitor characteristics include: number of stores in the business district, average price of stores in the same business district and the same category, etc.; store offline transaction characteristics include: maximum number of customers accepted by the store, etc.; competitor cooperation characteristics include: number of cooperative merchants in the same business district and the same category, discount rate of cooperative projects in the same business district and the same category, etc.; project characteristics include: project discount rate, average number of diners in the project, whether the project is available on holidays, total number of project reviews, project review score, project selling price, project market price, number of refunds on the previous day, project merchant share rate, etc.

[0037] In some embodiments of the present application, order transaction characteristics include: monthly store coupon verification volume, weekly store coupon verification volume, monthly store refund rate, weekly store refund rate, weekly store sales visit rate, order volume placed the day before the project, coupon verification volume the week before the project, etc.

[0038] Specifically in different application scenarios, the preset features may be different.

[0039] Then, the merchants who escaped the bill and the merchants who did not escape the bill are determined. The merchants who escaped the bill and the merchants who did not escape the bill can be merchants who escaped the bill or merchants who did not escape the bill determined based on user feedback or verification by business personnel.

[0040] Next, it is determined that the above-mentioned preset feature data of merchants who escape the bill constitute a positive sample set, and it is determined that the above-mentioned preset feature data of merchants who do not escape the bill constitute a negative sample set. In specific implementation, the feature values ​​of the above-mentioned preset features of merchants who escape the bill and merchants who do not escape the bill can be obtained respectively according to the historical order data, user data and merchant data of the platform. And the performance of the positive sample data in the positive sample set and the negative sample data in the negative sample set on the above-mentioned preset features is analyzed by the data analysis method in the prior art (such as using the linear correlation analysis method), and according to the analysis results, the preset number (such as 5) of the preset features with large performance differences in the positive sample data and the negative sample data are determined as the first features of escape from the bill. For example, the performance of the positive sample data and the negative sample data on the above-mentioned preset feature dimensions is analyzed, and it is found that the positive sample and the negative sample have the greatest performance differences in features such as the probability of refund, the frequency of online and offline visits to merchants and not purchasing, etc., then the two features of the probability of refund and the frequency of online and offline visits to merchants and not purchasing can be determined as the first features of escape from the bill.

[0041] Furthermore, according to each of the first features of escaping the bill, based on expert knowledge, auxiliary features associated with the first features of escaping the bill are selected from the preset features respectively, and the auxiliary features associated with the first features of escaping the bill and the first features of escaping the bill are used as descriptive features of the escaping behavior evidence item corresponding to the first features of escaping the bill. Usually, a single feature cannot objectively and accurately measure the merchant's escaping behavior, so other special provisions are needed for auxiliary explanation. For example, when the refund probability is used as the first feature of escaping the bill (that is, the main feature for measuring the merchant's escaping the bill), because the refund probability itself fluctuates greatly, the refund probability is meaningful only when the refund amount is considered. Therefore, the refund amount in the past 7 days, the refund amount in the past 7 days, etc. can be used as auxiliary features (that is, preset features associated with the first feature of escaping the bill).

[0042] Furthermore, it is also necessary to determine the characteristic value interval of the first escape feature and the characteristic value interval of the auxiliary feature associated with the first escape feature, which are used to determine the score of the evidence item and generate the evidence item. In some embodiments of the present application, for each of the first escape feature, determining the characteristic value interval of the escape behavior description feature includes: determining the characteristic value interval of the first escape feature based on the distribution data of the first escape feature value of merchants who have escaped the bill historically. For example, after respectively determining the refund probability of the store in the positive sample data and the negative sample data, the value range of the refund probability can be determined, such as 10% to 70%, then 10% to 70% is the characteristic value interval of the refund probability.

[0043] Furthermore, the preset feature thresholds of each feature describing the behavior of skipping a check are determined based on the experience of business experts. Taking the feature describing the behavior of skipping a check as an example, if the business expert believes that the merchant is in an abnormal state (i.e., suspected of skipping a check) when the refund probability is greater than 50%, then 50% is determined as the refund probability threshold.

[0044] In some embodiments of the present application, for the auxiliary feature associated with the first escape-the-bills feature, determining the feature value interval of the escape-the-bills behavior description feature includes: determining the feature value interval of the auxiliary feature according to expert knowledge.

[0045] Further, for each of the said escape behavior evidence items, a feature processing template matching the said escape behavior evidence items is constructed according to the feature value interval of each of the said escape behavior description features of the said escape behavior evidence items and the expert knowledge. The said feature processing template is used to indicate the judgment matching logic of the correspondence between the feature value of each escape behavior description feature of the said escape behavior evidence items and the feature value interval of the corresponding escape behavior description feature and the score that can generate the escape behavior evidence items. As mentioned above, by analyzing the historical data of the merchants of the platform, the feature interval value of the first escape behavior feature (i.e., the main feature) of the said escape behavior evidence items has been determined (such as 10% to 70%). Then, when constructing the feature processing template, the said feature interval value can be segmented according to the expert experience to obtain multiple sub-intervals, and a corresponding score can be set for each sub-interval. The score value is set according to the business experts and the business analysis results, and is adjusted according to the online operation of the feature processing template. The higher the score, the greater the probability of escape.

[0046] In some embodiments of the present disclosure, the feature processing template can be defined by a scripting language. In the feature processing template, it is necessary to define the scores corresponding to each feature value sub-interval of the escape behavior description feature as the main feature (i.e., the first escape feature, such as the refund probability), wherein the score is used to indicate that when the feature value of the escape behavior description feature is in the corresponding sub-interval, the escape behavior description feature can generate a score of the escape behavior evidence item. For example, when the 7-day refund probability is between 30% and 50%, the score is 5; when the 7-day refund probability is between 50% and 60%, the score is 6; when the 7-day refund probability is between 60% and 70%, the score is 7, and when it exceeds 70%, it is 8 points. In the feature processing template, it is also necessary to define the score corresponding to the feature value interval of the escape behavior description feature as an auxiliary feature (i.e., a preset feature associated with the main feature, such as: refund amount, refund amount), and the score corresponding to the auxiliary feature is used to generate the evidence item.

[0047] In some embodiments of the present disclosure, it also includes: constructing an evidence item generation template. The evidence item generation template is set in the form of variables and expressions, wherein at least some of the variables correspond to the score of the escape behavior description features that can generate the escape behavior evidence item, and the expression includes multiple fields, and each field is separated by "#". In some other embodiments of the present disclosure, the variables in the evidence item generation template can also represent information other than the escape behavior description features, such as merchant information, store information, etc. The evidence item generation template is used to generate evidence item description information according to the score of the escape behavior description features when the computing device executes the merchant escape behavior identification method disclosed in this application. In some embodiments of the present disclosure, the evidence item generation template is generated by a javascript script. The script content corresponding to each evidence item generation template is different. For example, a life template for an evidence item with a high probability of refund in a week can be expressed as:

[0048] Abnormality:

[0049] <script>if(${evidence#f#score#origin}>7){"高"}else if

[0050] (${evidence#f#score#origin}<＝7&&${evidence#f#score#origin}>＝6){"中"}else if(${evidence#f#score#origin}<6){"一般"}< / script>

[0051] Weekly refund volume: ${poi#f#refund7days#origin}, weekly orders:

[0052] ${poi#f#orderCnt7days#origin}, Weekly refund rate:

[0053] ${poi#f#refundRatio7days#origin}

[0054] The meaning of the above script is: the description information of the abnormality degree is: if the evidence score is greater than 7, it is replaced by the font "high"; if it is less than 7 and greater than 6, it is replaced by the font "medium", otherwise it is replaced by "general". At the same time, the evidence items also include: weekly refund volume, weekly order volume and weekly refund rate information.

[0055] In this embodiment, the refund probability is used as the first feature of the order escape, and the weekly refund volume, weekly order volume and weekly refund rate are used as auxiliary features. The values ​​of the refund probability, weekly refund volume, weekly order volume and weekly refund rate can be obtained through the transaction data of the platform. For example, the refund probability = 45%, the weekly refund volume = 80, the weekly order volume = 120, and the weekly refund rate = 20%.

[0056] In other embodiments of the present application, if multiple evidence items are defined in the expert model, the expert model also defines multiple groups of features describing the behavior of escaping the bill corresponding to the multiple evidence items, and then the values ​​of the multiple groups of features describing the behavior of escaping the bill need to be determined separately.

[0057] Step 120 , based on the expert model and the descriptive features of the at least one item of evidence of the escape behavior, respectively predict the escape probability of the target merchant corresponding to each item of evidence of the escape behavior.

[0058] In some embodiments of the present application, the step of predicting the probability of escape from the bill corresponding to each of the escape from the bill evidence items of the target merchant based on the expert model and the escape from the bill behavior description features of the at least one escape from the bill behavior evidence item includes: determining a feature processing template included in the expert model that matches the at least one escape from the bill behavior evidence item; using the feature processing template, respectively performing data processing on the feature values ​​of the escape from the bill behavior description features of the escape from the bill evidence items that match the feature processing template, and determining that each of the escape from the bill behavior description features of each of the escape from the bill behavior evidence items can generate a score for the escape from the bill behavior evidence item; for each of the escape from the bill behavior evidence items, determining the score of the preset escape from the bill description feature in each of the escape from the bill behavior description features of the escape from the bill behavior evidence item as the probability of escape from the bill corresponding to the escape from the bill evidence item of the target merchant. Preferably, the preset escape from the bill behavior description feature is the first escape from the bill feature. Because the first feature of running away from the bill is a feature with a large difference in performance between the positive sample data and the negative sample data, using the score of the first feature of running away from the bill as the probability of running away from the bill of the single evidence item of the target merchant can fully reflect the difference in the probabilities of running away from the bill of different merchants and avoid the problem of poor interpretability caused by using the scores of multiple features describing the running away from the bill as the probability of the merchant running away from the bill.

[0059] After determining the descriptive features of each escape behavior evidence item of the target merchant, data processing and judgment are performed on the descriptive features of the corresponding escape behavior evidence item based on the feature processing template corresponding to each escape behavior evidence item in the expert model to determine that each descriptive feature of the corresponding escape behavior evidence item can generate a score for the escape behavior evidence item.

[0060] In some embodiments of the present application, the step of performing data processing on the feature values ​​of the escape behavior description features of the escape behavior evidence items that match the feature processing template, and determining that each of the escape behavior description features of each of the escape behavior evidence items can generate a score for the escape behavior evidence item, includes: performing the following data processing on the feature values ​​of each of the escape behavior description features of the corresponding escape behavior evidence items by executing preset program instructions corresponding to the feature processing template: matching and judging the feature values ​​of the escape behavior description features with the corresponding relationship between preset feature value intervals and scores, and determining the score that matches the feature values ​​as the score for the escape behavior description features that can generate the score for the escape behavior evidence item; wherein the feature value intervals and the corresponding relationship between the feature value intervals and scores are determined based on historical escape behavior merchant data and an expert model. For example, the feature processing template configures the data processing flow of the "weekly refund rate" that describes the feature of the behavior of escaping the bill through a script: when the feature value of the "weekly refund rate" is between 30% and 50%, the corresponding score is 5 points; when the feature value of the "weekly refund rate" is between 50% and 60%, the corresponding score is 6 points; when the feature value of the "weekly refund rate" is between 60% and 70%, the corresponding score is 7 points; when the feature value of the "weekly refund rate" exceeds 70%, the corresponding score is 8 points. When the computing device executes the program instructions corresponding to the above data processing flow, the corresponding score can be obtained for the determined feature value of the "weekly refund rate".

[0061] The "weekly refund rate" feature describing the behavior of running away from the bill is a feature describing the behavior of running away from the bill in the expert model that corresponds to the "refund probability" item of evidence for the behavior of running away from the bill. By executing the program instructions corresponding to all feature processing flows configured in the feature processing template that matches the "refund probability" item of evidence for the behavior of running away from the bill, the score of the feature value of each feature describing the behavior of running away from the bill of the "refund probability" item of evidence for the behavior of running away from the bill can be obtained. Finally, the score of each feature value is used as the corresponding feature describing the behavior of running away from the bill to generate the score of the evidence item for the behavior of running away from the bill.

[0062] In some embodiments of the present application, the score of the first escape feature in the escape behavior description features of the escape behavior evidence item is used as the escape probability of the target merchant based on the escape behavior evidence item.

[0063] In other embodiments of the present application, if the expert template is configured with multiple evidence items for identifying merchants' escape behavior, it is necessary to execute the feature processing flow in the feature processing template that matches each evidence item separately to determine the scores corresponding to each of the multiple groups of escape behavior description features, that is, to determine the scores of the escape behavior evidence items generated by each of the multiple groups of escape behavior description features.

[0064] In some embodiments of the present application, after the step of performing data processing on the feature values ​​of the escape behavior description features of the escape behavior evidence items that match the feature processing template respectively through the feature processing template, and determining that each of the escape behavior description features of each of the escape behavior evidence items can generate a score for the escape behavior evidence item, the step further includes: for each of the escape behavior evidence items, generating descriptive information of the escape behavior evidence item based on the determined score for the escape behavior evidence item can be generated for each of the escape behavior description features of the escape behavior evidence item and the evidence item generation template included in the expert model.

[0065] Furthermore, after determining that each of the multiple groups of escape behavior description features can generate a score for the escape behavior evidence item, the instructions corresponding to the evidence item generation template in the expert template are further executed to generate the description information of the corresponding escape behavior evidence item according to each group of escape behavior description features. In some embodiments of the present application, the description information of the escape behavior evidence item includes the escape behavior description features of the escape behavior evidence item and its characteristic values ​​and / or scores, grades and other information. For example, the description information of the evidence item of high refunds caused by Zhou's suspected escape behavior is: "The abnormal degree of high refunds caused by Zhou's suspected escape behavior: The number of refunds caused by Zhou's suspected escape behavior: 1 The number of refund orders in seven days: 6 The refund rate in seven days: 0.34". By generating the description information of the escape behavior evidence item and displaying the basis for identifying the escape behavior in the description information of the escape behavior evidence item, the identification of the escape behavior is made more explainable.

[0066] In some embodiments of the present application, in order to improve the efficiency of identifying escape-the-bills behavior, before performing data processing on the feature values ​​of the escape-the-bills behavior description features of the escape-the-bills behavior evidence items that match the feature processing template through the feature processing template, and determining that each of the escape-the-bills behavior description features of each of the escape-the-bills behavior evidence items can generate a score for the escape-the-bills behavior evidence item, the method also includes: determining whether the escape-the-bills behavior description features of each of the escape-the-bills behavior evidence items meet a preset feature threshold; if so, performing data processing on the feature values ​​of the escape-the-bills behavior description features of the escape-the-bills behavior evidence items that match the feature processing template through the feature processing template, and determining that each of the escape-the-bills behavior description features of each of the escape-the-bills behavior evidence items can generate a score for the escape-the-bills behavior evidence item.

[0067] Among them, the preset feature threshold is determined based on the analysis results of the data stored in the platform and the expert experience when constructing the feature processing template. The preset feature threshold can be set in the feature processing template, or it can be stored through an independent data file. For example, in the feature processing template, the feature threshold of each escape behavior description feature of the escape behavior evidence item matching the feature processing template can be pre-set using a scripting language. Then, when identifying the merchant's escape behavior, for all the escape behavior description features of each escape behavior evidence item, the feature value of the escape behavior description feature is compared with the feature threshold of the escape behavior description feature. If the feature values ​​of all the escape behavior description features of a certain escape behavior evidence item are greater than or equal to the feature threshold, it is determined that the certain escape behavior evidence item is established, and the subsequent description information generation step of the escape behavior evidence item can be continued. Otherwise, the subsequent data processing based on the configuration of the feature processing template is not performed to improve the data processing efficiency. In some embodiments of the present application, only when the escape behavior evidence item is established can it be determined that the merchant has escaped the bill.

[0068] Step 130 , determining a first bill-avoidance probability of the target merchant by linearly fusing the bill-avoidance probabilities corresponding to each of the bill-avoidance behavior evidence items of the target merchant.

[0069] In some embodiments of the present application, if a certain item of evidence of escape behavior configured by the expert model is established, the probability of escape of a target merchant based on the item of evidence of escape behavior can be determined based on the expert model. Therefore, based on the expert model, the probability of escape of a target merchant corresponding to 0 to multiple items of evidence of escape behavior can be determined.

[0070] In some embodiments of the present application, before linearly fusing the escape probability corresponding to each of the escape behavior evidence items of the target merchant, it is usually necessary to normalize the escape probability of each of the escape behavior evidence items to unify the dimensions.

[0071] Furthermore, by linearly fusing the escape probabilities corresponding to the escape behavior evidence items of the target merchant, the fusion result is determined as the first escape probability of the target merchant. For example, the normalized escape probabilities corresponding to the escape behavior evidence items of the target merchant are added or averaged, and the result is used as the first escape probability of the target merchant.

[0072] Step 140, based on a plurality of preset machine learning models, respectively predict the probability of skipping the bill for the target merchant, and obtain a second skipping probability corresponding to each of the machine learning models.

[0073] In some embodiments of the present application, multiple machine learning models can be preset, and the probability of the target merchant running away from the bill can be predicted by each of the machine learning models. The multiple machine learning models have differences in the selection range of positive and negative samples in the training data, the selection of descriptive features of the running away from the bill behavior, and the model algorithm dimensions. The multiple machine learning models can be XGBoost models or linear models or other models in the prior art.

[0074] The training process of the multiple machine learning models can be referred to in the prior art and will not be described in detail in the embodiments of the present application.

[0075] In the process of identifying the behavior of a merchant who escapes the bill, for each of the machine learning models, the following operations are performed respectively: the characteristic values ​​of the description features of the behavior of escaping the bill that match the machine learning model are determined respectively according to the data associated with the merchant; the characteristic values ​​of the description features of the behavior of escaping the bill are input into the machine learning model to obtain the second probability of escaping the bill corresponding to the machine learning model. At this point, the second probability of escaping the bill that is output by each of the machine learning models after predicting the behavior of escaping the bill for the target merchant can be obtained.

[0076] Step 150, according to the first probability of escaping the bill and the second probability of escaping the bill corresponding to each of the machine learning models, the probability of escaping the bill of the target merchant is calculated by a preset fusion model.

[0077] After the aforementioned steps, a first probability of the target merchant running away from the bill based on the expert model and multiple second probabilities of running away from the bill can be determined, wherein each of the second probabilities of running away from the bill corresponds to one of the machine learning models.

[0078] In some embodiments of the present application, the step of calculating the probability of escape from the order of the target merchant through a preset fusion model based on the first probability of escape from the order and the second probability of escape from the order corresponding to each of the machine learning models includes: performing a weighted sum operation on the second probability of escape from the order corresponding to each of the machine learning models to obtain a third probability of escape from the order; performing a weighted fusion operation on the first probability of escape from the order and the third probability of escape from the order to obtain the probability of escape from the order of the target merchant. Wherein, when performing a weighted sum operation on the second probability of escape from the order corresponding to each of the machine learning models, the weighted values ​​of the second probability of escape from the order corresponding to different machine learning models are determined by training the fusion model; when performing a weighted fusion operation on the first probability of escape from the order and the third probability of escape from the order, the weighted values ​​of the first probability of escape from the order and the third probability of escape from the order are determined by training the fusion model.

[0079] Take the preset fusion model as an example:

[0080] Among them, escape is the target merchant identifier, Pexpert_model (escape) is the first escape probability, W1 is the weight of the first escape probability, is the second escape probability predicted by machine learning model i for the target merchant escape, W2 is the weight of the third escape probability, n matches the number of machine learning models, and n is a natural number greater than or equal to 1. The two parameter values ​​of W1 and W2 determine the influence ratio of the expert model and multiple machine learning models on the escape probability.

[0081] Through the above fusion model, the probability of each target merchant running away from the bill can be obtained. Since the first probability of running away from the bill is obtained through the expert model, the expert model predicts the elimination probability based on prior knowledge, and there is no cold start problem. Even if there is no data of merchants running away from the bill reported by users or business personnel, the merchant's running away from the bill behavior can still be predicted. The machine learning model can mine the potential data distribution relationship between merchant data, user data and platform transaction data, and there is no strong constraint. Therefore, according to the above fusion model, it can be seen that the use of the above fusion model to predict the merchant's running away from the bill combines prior knowledge and real-time data, which is conducive to improving the accuracy of the merchant's running away from the bill prediction.

[0082] The embodiment of the present application discloses a method for identifying a merchant's escape behavior, which comprises determining the escape behavior description features of at least one escape behavior evidence item of a target merchant based on an expert model; predicting the escape probability of the target merchant corresponding to each escape behavior evidence item based on the expert model and the escape behavior description features of the at least one escape behavior evidence item; determining a first escape probability of the target merchant by linearly fusing the escape probabilities corresponding to each escape behavior evidence item of the target merchant; predicting the escape probability of the target merchant based on a plurality of preset machine learning models to obtain second escape probabilities corresponding to each machine learning model; and calculating the escape probability of the target merchant through a preset fusion model based on the first escape probability and the second escape probability corresponding to each machine learning model, thereby solving the problem of at least low recognition accuracy in the prior art method for identifying a merchant's escape behavior.

[0083] Furthermore, when predicting the first escape probability of the target merchant based on the expert model, the escape behavior description feature of at least one escape behavior evidence item is combined, so that the determined escape probability of the target merchant is more interpretable.

[0084] Embodiment 2

[0085] Based on the first embodiment, a method for identifying a merchant's behavior of escaping from a purchase order is disclosed in this embodiment, such as Figure 2As shown, before the step of calculating the probability of the target merchant escaping the bill by a preset fusion model based on the first probability of escaping the bill and the second probability of escaping the bill corresponding to each of the machine learning models, it further includes:

[0086] Step 160, determine whether there is any evidence item of the behavior of skipping the bill. If so, execute step 150, otherwise, output the failure of predicting the probability of skipping the bill of the target merchant.

[0087] In some preferred embodiments of the present application, before the step of calculating the escape probability of the target merchant through a preset fusion model based on the first escape probability and the second escape probability corresponding to the machine learning model, it is first determined whether there is any evidence item of escape behavior. Only if there is evidence item of escape behavior, the step of fusion to determine the escape probability is performed. Otherwise, the output predicting the escape probability of the target merchant fails, and the escape probability calculation of the current merchant is terminated.

[0088] In some embodiments of the present application, it can be determined whether there is an item of evidence for running away from the order based on the comparison result of the preset feature threshold of the description feature of the behavior for running away from the order of the evidence item. If the feature values ​​of the description feature of the behavior for running away from the order of a certain evidence item for whether running away from the order is an order all meet the corresponding preset feature threshold, then it is determined that the item of evidence for running away from the order exists. Taking the evidence item of Zhou's suspected running away from the order causing a high refund rate in Example 1 as an example, if the description features of the behavior for running away from the order of the evidence item: refund probability, weekly refund amount, weekly order volume and weekly refund rate all meet the corresponding preset feature threshold, then it is determined that there is an item of evidence for Zhou's suspected running away from the order causing a high refund rate; if the description features of the behavior for running away from the order of the evidence item: refund probability, weekly refund amount and weekly order volume all meet the corresponding preset feature threshold, but the weekly refund rate does not meet the corresponding preset feature threshold, then it is determined that the item of evidence for Zhou's suspected running away from the order causing a high refund rate does not hold.

[0089] When it is identified that a merchant has escaped the bill, the prior art cannot provide strong evidence, which often leads to the merchant not recognizing the platform's judgment result or the platform's judgment result being wrong. The embodiment of the present application further determines whether the merchant has escaped the bill based on the probability of escape after determining that there is evidence of the merchant's escape. When it is determined that the merchant has escaped the bill, the corresponding evidence item can be presented to avoid the merchant's disapproval of the platform's judgment result. At the same time, the evidence item can also be used by the platform to manually verify the judgment result, further improving the accuracy of the identification of escape behavior.

[0090] In other embodiments of the present application, before the step of calculating the probability of escape from the bill of the target merchant through a preset fusion model based on the first probability of escape from the bill and the second probability of escape from the bill corresponding to the machine learning model, it also includes: filtering and judging the target merchant based on the escape from the bill processing record of the target merchant, a preset list of excluded merchants, and the evidence items of the escape from the bill corresponding to the target merchant.

[0091] Among them, the skipping bill processing record is determined according to the skipping bill behavior identification result, or according to the settings of the business personnel. For example, for the target merchant identified as a skipping bill merchant by the skipping bill probability predicted by the expert model, first determine whether the merchant is included in the skipping bill processing record stored by the platform for a preset time (such as within 1 month). If it is in the skipping bill processing record, it means that the merchant has been determined to be a skipping bill merchant, and there is no need to combine the processing results of other models for identification, and the target merchant is screened out from the merchants that need to be identified by the fusion model; otherwise, it is necessary to combine the skipping bill probability predicted by other models (such as machine learning models) for fusion identification.

[0092] The preset excluded merchant list is set according to business needs. The merchants in the preset excluded merchant list are merchants whose behavior of skipping the order should not be identified from the business point of view, such as strategic big customers.

[0093] On the other hand, for each merchant that the machine learning model predicts to have engaged in check-fleeing behavior, it is necessary to screen out those merchants that do not have evidence of check-fleeing behavior.

[0094] By filtering merchants, the number of merchants for fusion identification can be reduced, the amount of data processing by computing devices can be reduced, and the computational efficiency of identifying skipping orders can be improved.

[0095] Embodiment 3

[0096] This embodiment discloses a device for identifying a merchant's behavior of skipping a payment. Figure 3 As shown, the device comprises:

[0097] A skipping behavior description feature determination module 310 is used to determine a skipping behavior description feature of at least one skipping behavior evidence item of a target merchant based on an expert model;

[0098] A single evidence item escape probability determination module 320, configured to predict the escape probability of the target merchant corresponding to each escape behavior evidence item based on the expert model and the escape behavior description feature of the at least one escape behavior evidence item;

[0099] A first skipping probability determination module 330, configured to determine a first skipping probability of the target merchant by linearly fusing the skipping probabilities corresponding to each of the skipping behavior evidence items of the target merchant;

[0100] A second skipping probability determination module 340, configured to predict the skipping probability of the target merchant based on a plurality of preset machine learning models, and obtain second skipping probabilities corresponding to the machine learning models;

[0101] The bill-avoidance probability fusion module 350 is used to calculate the bill-avoidance probability of the target merchant through a preset fusion model based on the first bill-avoidance probability and the second bill-avoidance probability corresponding to each of the machine learning models.

[0102] In some embodiments of the present application, the single evidence item escape probability determination module 320 is further used to:

[0103] Determining a feature processing template included in the expert model that matches the at least one item of evidence of skipping the bill;

[0104] By using the feature processing template, data processing is performed on the feature values ​​of the fleeing behavior description features of the fleeing behavior evidence items that match the feature processing template, and it is determined that each fleeing behavior description feature of each fleeing behavior evidence item can generate a score for the fleeing behavior evidence item;

[0105] For each of the items of evidence of escape behavior, the scores of the preset escape behavior description features in each of the escape behavior description features of the item of evidence of escape behavior are determined as the escape probability of the target merchant corresponding to the item of evidence of escape behavior.

[0106] Because the preset features describing the behavior of running away from the bill are features that show significant differences in performance between the positive sample data and the negative sample data, using the score of the preset features describing the behavior of running away from the bill as the probability of running away from the bill of the single evidence item of the target merchant's behavior of running away from the bill can fully reflect the differences in the probabilities of running away from the bill of different merchants and avoid the problem of poor interpretability caused by using the scores of multiple features describing the behavior of running away from the bill as the probability of a merchant running away from the bill.

[0107] Optionally, the step of performing data processing on the feature values ​​of the escape behavior description features of the escape behavior evidence items matching the feature processing template respectively through the feature processing template, and determining that each of the escape behavior description features of each of the escape behavior evidence items can generate a score for the escape behavior evidence item, includes:

[0108] By executing the preset program instructions corresponding to the feature processing template, the feature values ​​of the respective features describing the escape behavior of the corresponding escape behavior evidence items are respectively subjected to the following data processing:

[0109] Matching and judging the characteristic value of the feature describing the behavior of fleeing the bill with the corresponding relationship between the preset characteristic value interval and the score, and determining the score matching the characteristic value as the score of the evidence item of the behavior of fleeing the bill that can be generated by the feature describing the behavior of fleeing the bill;

[0110] The characteristic value interval and the corresponding relationship between the characteristic value interval and the score are determined based on historical data of merchants who escaped the order and an expert model.

[0111] Optionally, after the step of performing data processing on the feature values ​​of the escape behavior description features of the escape behavior evidence items matching the feature processing template respectively through the feature processing template and determining that each of the escape behavior description features of each of the escape behavior evidence items can generate a score for the escape behavior evidence item, the step further includes:

[0112] For each of the items of evidence of fleeing the bill, a score of the item of evidence of fleeing the bill and an evidence item generation template included in the expert model can be generated based on the determined description features of the item of evidence of fleeing the bill, and descriptive information of the item of evidence of fleeing the bill can be generated.

[0113] Optionally, before the step of performing data processing on the feature values ​​of the escape behavior description features of the escape behavior evidence items matching the feature processing template respectively by using the feature processing template to determine that each of the escape behavior description features of each of the escape behavior evidence items can generate a score for the escape behavior evidence item, the step further includes:

[0114] Determine respectively whether the descriptive features of the escape behavior of each of the escape behavior evidence items meet a preset feature threshold;

[0115] If the conditions are met, the feature processing template is used to perform data processing on the feature values ​​of the escape behavior description features of the escape behavior evidence items that match the feature processing template, and it is determined that the escape behavior description features of each escape behavior evidence item can generate a score for the escape behavior evidence item.

[0116] In some embodiments of the present application, see Figure 4 The device further comprises: a feature processing template construction module 300, wherein the feature processing template construction module 300 is used to:

[0117] Obtaining preset characteristics of a number of merchants who have skipped bills and the preset characteristics of a number of merchants who have not skipped bills;

[0118] According to the analysis results of the performance of the merchant who escaped the bill and the merchant who did not escape the bill on each of the preset features, a preset number of preset features with relatively large performance differences between the merchant who escaped the bill and the merchant who did not escape the bill are determined as first features of escaped the bill;

[0119] For each of the first features of escaping the bill, respectively determine the preset features associated with the first feature of escaping the bill, and use the preset features associated with the first feature of escaping the bill and the first feature of escaping the bill as escaping behavior description features of the escaping behavior evidence item corresponding to the first feature of escaping the bill;

[0120] For each of the first features of escaping the bill, respectively determine a feature value interval and a feature threshold of each of the features describing the escaping the bill behavior of the item of evidence corresponding to the first feature of escaping the bill;

[0121] For each of the items of evidence of the escape behavior, a feature processing template matching the item of evidence of the escape behavior is constructed according to a feature value interval of each feature describing the escape behavior of the item of evidence of the escape behavior and expert knowledge.

[0122] In some embodiments of the present application, the preset characteristics include: merchant characteristics, user characteristics, and order transaction characteristics.

[0123] In some embodiments of the present application, see Figure 4 The device also includes: a merchant filtering module 360. Before calculating the escape probability of the target merchant through a preset fusion model based on the first escape probability and the second escape probability corresponding to the machine learning model, the merchant filtering module 360 ​​is further used to: filter and judge the target merchant based on the escape processing record of the target merchant, the preset list of excluded merchants, and the escape behavior evidence item corresponding to the target merchant.

[0124] In some embodiments of the present application, see Figure 4 The device also includes an evidence item judgment module 370, and the evidence item judgment module 370 is used to determine whether there is an evidence item of escape from the bill; the evidence item judgment module 370 is also used to execute the step of calculating the escape probability of the target merchant through a preset fusion model based on the first escape probability and the second escape probability corresponding to the machine learning model if there is an evidence item of escape from the bill, otherwise, outputting a failure to predict the escape probability of the target merchant.

[0125] The device for identifying merchant's behavior of running away from the bill disclosed in the embodiment of the present application is used to implement the steps of the method for identifying merchant's behavior of running away from the bill described in Embodiment 1 and Embodiment 2 of the present application. The specific implementation methods of each module of the device refer to the corresponding steps and will not be repeated here.

[0126] The device for identifying merchant escape behavior disclosed in the embodiment of the present application determines the escape behavior description features of at least one escape behavior evidence item of a target merchant based on an expert model; predicts the escape probability of the target merchant corresponding to each escape behavior evidence item based on the expert model and the escape behavior description features of the at least one escape behavior evidence item; determines the first escape probability of the target merchant by linearly fusing the escape probabilities corresponding to each escape behavior evidence item of the target merchant; predicts the escape probability of the target merchant based on preset multiple machine learning models to obtain second escape probabilities corresponding to each machine learning model; calculates the escape probability of the target merchant through a preset fusion model based on the first escape probability and the second escape probability corresponding to each machine learning model, thereby solving the problem of at least low recognition accuracy in the merchant escape behavior identification method in the prior art.

[0127] Furthermore, when predicting the first escape probability of the target merchant based on the expert model, the escape behavior description feature of at least one escape behavior evidence item is combined, so that the determined escape probability of the target merchant is more interpretable.

[0128] Correspondingly, the present application also discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for identifying the merchant's behavior of escaping the order as described in Embodiments 1 and 2 of the present application is implemented. The electronic device may be a PC, a mobile terminal, a personal digital assistant, a tablet computer, etc.

[0129] The present application also discloses a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the method for identifying the merchant's behavior of escaping the payment as described in Embodiment 1 and Embodiment 2 of the present application are implemented.

[0130] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0131] The above is a detailed introduction to a method and device for identifying merchant escape from payment provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

[0132] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

Claims

1. A method for identifying a merchant's behavior of skipping a payment, characterized in that: include: Determine, based on the expert model, a descriptive feature of at least one evidence item of the target merchant's behavior of running away from the bill; Based on the expert model and the description features of the at least one item of evidence of the escape behavior, respectively predict the escape probability of the target merchant corresponding to each item of evidence of the escape behavior; Determine a first probability of skipping a bill for the target merchant by linearly fusing the skipping bill probabilities corresponding to each of the skipping bill behavior evidence items of the target merchant; Based on a plurality of preset machine learning models, respectively predict the probability of skipping the order for the target merchant, and obtain a second skipping probability corresponding to each of the machine learning models; Calculating the probability of the target merchant escaping the order through a preset fusion model according to the first probability of escaping the order and the second probability of escaping the order corresponding to each of the machine learning models; The step of predicting the probability of the target merchant of escaping the bill corresponding to each of the escaping the bill evidence items based on the expert model and the escaping the bill behavior description features of the at least one escaping the bill behavior evidence item comprises: Determining a feature processing template included in the expert model that matches the at least one item of evidence of skipping the bill; By using the feature processing template, data processing is performed on the feature values ​​of the fleeing behavior description features of the fleeing behavior evidence items that match the feature processing template, and it is determined that each fleeing behavior description feature of each fleeing behavior evidence item can generate a score for the fleeing behavior evidence item; For each of the items of evidence of escape behavior, determining the score of the preset escape behavior description feature in each of the escape behavior description features of the item of evidence of escape behavior as the escape probability of the target merchant corresponding to the item of evidence of escape behavior; The step of performing data processing on the feature values ​​of the escape behavior description features of the escape behavior evidence items that match the feature processing template respectively through the feature processing template, and determining that each of the escape behavior description features of each of the escape behavior evidence items can generate a score for the escape behavior evidence item, comprises: By executing the preset program instructions corresponding to the feature processing template, the feature values ​​of the respective features describing the escape behavior of the corresponding escape behavior evidence items are respectively subjected to the following data processing: Matching and judging the characteristic value of the feature describing the behavior of fleeing the bill with the corresponding relationship between the preset characteristic value interval and the score, and determining the score matching the characteristic value as the score of the evidence item of the behavior of fleeing the bill that can be generated by the feature describing the behavior of fleeing the bill; The characteristic value interval and the corresponding relationship between the characteristic value interval and the score are determined based on historical data of merchants who escaped from the order and an expert model; Before the step of determining the descriptive features of at least one item of evidence of the target merchant's escape behavior based on the expert model, the method further includes: Obtaining preset characteristics of a number of merchants who have skipped bills and the preset characteristics of a number of merchants who have not skipped bills; According to the analysis results of the performance of the merchant who escaped the bill and the merchant who did not escape the bill on each of the preset features, a preset number of preset features with relatively large performance differences between the merchant who escaped the bill and the merchant who did not escape the bill are determined as first features of escaped the bill; For each of the first features of escaping the bill, respectively determine the preset features associated with the first feature of escaping the bill, and use the preset features associated with the first feature of escaping the bill and the first feature of escaping the bill as escaping behavior description features of the escaping behavior evidence item corresponding to the first feature of escaping the bill; For each of the first features of escaping the bill, respectively determine a feature value interval and a feature threshold of each of the features describing the escaping the bill behavior of the item of evidence corresponding to the first feature of escaping the bill; For each of the items of evidence of the escape behavior, a feature processing template matching the item of evidence of the escape behavior is constructed according to a feature value interval of each feature describing the escape behavior of the item of evidence of the escape behavior and expert knowledge.

2. The method according to claim 1, characterized in that Before the step of performing data processing on the feature values ​​of the escape behavior description features of the escape behavior evidence items that match the feature processing template respectively using the feature processing template and determining that each of the escape behavior description features of each of the escape behavior evidence items can generate a score for the escape behavior evidence item, the step further includes: Determine respectively whether the descriptive features of the escape behavior of each of the escape behavior evidence items meet a preset feature threshold; If the conditions are met, the feature processing template is used to perform data processing on the feature values ​​of the escape behavior description features of the escape behavior evidence items that match the feature processing template, and it is determined that the escape behavior description features of each escape behavior evidence item can generate a score for the escape behavior evidence item.

3. The method according to claim 1, characterized in that The preset features include: merchant features, user features, and order transaction features.

4. The method according to any one of claims 1 to 3, characterized in that: Before the step of calculating the probability of escape from the bill of the target merchant by a preset fusion model according to the first probability of escape from the bill and the second probability of escape from the bill corresponding to each of the machine learning models, the method further includes: The target merchant is filtered and judged based on the target merchant's skipping payment processing record, the preset excluded merchant list, and the skipping payment behavior evidence item corresponding to the target merchant.

5. The method according to claim 1, characterized in that Before the step of calculating the probability of escape from the bill of the target merchant by a preset fusion model according to the first probability of escape from the bill and the second probability of escape from the bill corresponding to each of the machine learning models, the method further includes: Determine whether there is evidence of skipping the bill; If so, execute the step of calculating the probability of the target merchant running away from the bill through a preset fusion model based on the first probability of running away from the bill and the second probability of running away from the bill corresponding to the machine learning model; otherwise, output the prediction of the probability of the target merchant running away from the bill failing.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for identifying the merchant's behavior of running away from the bill as described in any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for identifying merchant's behavior of skipping the payment as described in any one of claims 1 to 5 are implemented.

Images