A payment reminder information determination method, model training method and parking fee management system

Abstract

The embodiments of the present application provide a method for determining collection information, a model training method and a parking fee management system, which relate to the field of artificial intelligence technology. The method comprises: obtaining historical parking data of each candidate user who has arrears in a parking lot within a historical time period; for each candidate user, determining parking feature data of the candidate user based on the historical parking data of the candidate user; the parking feature data comprises: parking behavior characteristics of the candidate user, and / or, collection sensitivity index; processing the parking feature data of each candidate user based on a pre-trained collection prediction model, and obtaining target users recommended for collection among each candidate user based on the output result of the collection prediction model, which can improve the accuracy of determining the target users for collection, and thus improve the success rate of collection.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method for determining payment information, a model training method, and a parking fee management system. Background Art

[0002] For open parking lots, such as roadside parking lots, there are only designated parking spaces and no gates to record vehicle entry and exit, so users may forget to pay after leaving the parking lot. Therefore, users who owe fees need to be urged to pay.

[0003] In the related art, it is necessary to manually select the target users for collection from the users who owe payment and send collection SMS messages. However, it is time-consuming and laborious to edit a SMS message for each user who owes payment. If collection is collected too frequently, the SMS or traffic cost will be high, and the user may even be blocked. If the collection is collected manually, it is easy to make mistakes and the collection success rate cannot be guaranteed. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a method for determining payment information, a model training method and a parking fee management system, so as to improve the accuracy of determining the target users for payment, thereby improving the success rate of payment. The specific technical solution is as follows:

[0005] In a first aspect, in order to achieve the above-mentioned purpose, an embodiment of the present application provides a method for determining collection information, the method comprising:

[0006] Obtain historical parking data of each candidate user in a historical time period; wherein the candidate user is a user who has outstanding fees in a parking lot; for each candidate user, determine parking feature data of the candidate user based on the historical parking data of the candidate user; wherein the parking feature data includes: parking behavior characteristics of the candidate user, and / or a collection sensitivity index; process the parking feature data of each candidate user based on a pre-trained collection prediction model, and obtain a target user recommended for collection among each candidate user based on an output result of the collection prediction model.

[0007] Optionally, the determining, for each candidate user, parking characteristic data of the candidate user based on the historical parking data of the candidate user includes: determining, for each candidate user, parking behavior characteristics, a credit index, and a collection sensitivity index of the candidate user based on the historical parking data of the candidate user;

[0008] The collection prediction model based on pre-training processes the parking feature data of each candidate user, and obtains the target user recommended for collection among each candidate user based on the output result of the collection prediction model, including: inputting the parking behavior characteristics, credit index and collection sensitivity index of each candidate user into the pre-trained collection prediction model, and obtaining the collection success rate of each candidate user output by the collection prediction model; wherein the collection prediction model is: trained based on the parking behavior characteristics, credit index and collection sensitivity index of each sample user, and sample labels representing the historical number of successful collections of each sample user; based on the collection success rate of each candidate user, determining the target user recommended for collection from each candidate user.

[0009] Optionally, the collection prediction model based on pre-trained processing the parking feature data of each candidate user, and based on the output result of the collection prediction model, obtains the target user recommended for collection among each candidate user, including: inputting the parking feature data of each candidate user into the pre-trained collection prediction model, and obtaining the target user recommended for collection output by the collection prediction model; wherein the collection prediction model is obtained by training based on the parking feature data of each sample user and a sample label indicating whether the collection for each sample user is successful.

[0010] Optionally, after determining the parking behavior characteristics, credit index and collection sensitivity index of each candidate user based on the historical parking data of the candidate user, the method further includes: grouping the candidate users based on the parking behavior characteristics of the candidate users to obtain a plurality of candidate user groups;

[0011] The step of inputting the parking behavior characteristics, credit index and collection sensitivity index of each candidate user into a pre-trained collection prediction model to obtain the collection success rate of each candidate user output by the collection prediction model comprises: for each candidate user group, inputting the parking behavior characteristics, credit index and collection sensitivity index of the group of candidate users into the collection prediction model corresponding to the candidate user group to obtain the collection success rate of the group of candidate users output by the collection prediction model; wherein the collection prediction model corresponding to the candidate user group is obtained by training based on the parking behavior characteristics, credit index and collection sensitivity index of each sample user of the corresponding sample user group, and sample labels representing the number of historical collection successes of each sample user of the sample user group;

[0012] The step of determining the target user recommended for payment collection from each candidate user based on the payment collection success rate of each candidate user includes: determining the target user recommended for payment collection from each group of candidate users based on the payment collection success rate of each group of candidate users.

[0013] Optionally, the method of determining target users recommended for collection from each group of candidate users based on the collection success rate of each group of candidate users includes: for each candidate user group, determining the number of recommended collection users in the group of candidate users as a first number based on the number of candidate users in the group and the prediction accuracy of the collection prediction model corresponding to the candidate user group for different numbers of users; wherein the prediction accuracy is: determined by predicting the collection success rate of the corresponding test user group based on the collection prediction model corresponding to the candidate user group; determining the first first number of candidate users from the group of candidate users according to the target sorting result of the collection success rate of the group of candidate users from high to low, and obtaining the recommended collection users in the candidate user group; obtaining the recommended collection users in each candidate user group to obtain the target users recommended for collection.

[0014] Optionally, the method for determining the prediction accuracy of the collection prediction model corresponding to the candidate user group for different numbers of users includes: for each test user group, inputting the parking behavior characteristics, credit index and collection sensitivity index of the group of test users into the corresponding collection prediction model to obtain the collection success rate of the group of test users; sorting the collection success rate of the group of test users from high to low to obtain a first test sorting result for the group of test users; sorting the group of test users from high to low according to the historical number of successful collections as the second test sorting result for the group of test users; comparing the first second number of test users in the second test sorting result with the first second number of test users in the first test sorting result to determine the prediction accuracy of the collection prediction model corresponding to the candidate user group corresponding to the test user group for the test user group.

[0015] Optionally, the parking behavior characteristics include: the candidate user's historical arrears amount and the number of historical successful collection requests;

[0016] The method groups the candidate users based on the parking behavior characteristics of the candidate users to obtain multiple candidate user groups, including: grouping the candidate users whose historical arrears amounts belong to the same amount range and whose historical successful collection times belong to the same number range into the same candidate user group; wherein different candidate user groups correspond to different amount ranges and number ranges; and the candidate user group and the corresponding sample user group have historical arrears amounts that belong to the same amount range and historical successful collection times that belong to the same number range.

[0017] Optionally, for each candidate user, based on the historical parking data of the candidate user, determining the parking behavior characteristics, credit index and collection sensitivity index of the candidate user, including: for each candidate user, counting the historical parking data of the candidate user to obtain the parking behavior characteristics of the candidate user; calculating the payment deferral index of the candidate user based on the departure time, payment due time and actual payment time of each parking behavior of the candidate user, and calculating the credit index of the candidate user based on the payment deferral index of the candidate user; when the candidate user is not collected for payment, determining the collection sensitivity index of the candidate user to be a first value; when the historical number of successful collections of the candidate user is 0, determining the collection sensitivity index of the candidate user to be a second value; when the historical number of successful collections of the candidate user is greater than 0, calculating the collection sensitivity index of the candidate user based on the actual payment time, collection time and preset payment due time of each parking behavior of the candidate user that is collected; wherein, the preset payment due time refers to the time between the departure time of the vehicle and the payment due time.

[0018] Optionally, after determining the target users recommended for collection from the candidate users based on the collection success rate of each candidate user, the method further includes: dividing the target users whose collection success rate belongs to the same success rate range into a target user group to obtain multiple target user groups; for each target user group, determining the collection frequency of the target users in the group based on the success rate range to which the collection success rate of the target users in the group belongs, and the corresponding relationship between the preset success rate range and the collection frequency; and collecting collection from the target users in the group according to the collection frequency of the target users in the group.

[0019] Optionally, the historical parking data includes at least one of the following: the candidate user's vehicle entry data, vehicle exit data, payment data and collection data; the vehicle entry data includes: parking lot identification and type; the vehicle exit data includes: parking duration and order fee; the payment data includes: payment method and payment type; the collection data includes: collection amount and collection amount.

[0020] Optionally, the collection prediction model is an extreme gradient boosting XGBoost model; or a gradient boosting decision tree GBDT model.

[0021] In a second aspect, in order to achieve the above-mentioned purpose, an embodiment of the present application provides a model training method, the method comprising:

[0022] Obtain historical parking data of each sample user within a historical time period; generate training samples based on the historical parking data of each sample user; wherein the training samples include: parking behavior characteristics, credit index and collection sensitivity index of each sample user, and sample labels indicating the number of historical collection successes of each sample user; input the parking behavior characteristics, credit index and collection sensitivity index of each sample user into the collection prediction model of the initial structure to obtain the collection success rate of each sample user; based on the collection success rate of each sample user and the sample labels indicating the number of historical collection successes of each sample user, adjust the model parameters of the collection prediction model of the initial structure to obtain a trained collection prediction model.

[0023] Optionally, the generating of training samples based on the historical parking data of each sample user includes: for each sample user, counting the historical parking data of the sample user to obtain the parking behavior characteristics of the sample user, and calculating the credit index and collection sensitivity index of the sample user based on the historical parking data of the sample user; for every two sample users, if the historical number of successful collections of the first sample user of the two sample users is greater than the historical number of successful collections of the second sample user, a positive sample is generated; wherein the positive sample includes the parking behavior characteristics, credit index and collection sensitivity index of the two sample users, and a first sample label indicating the historical number of successful collections of the two sample users; if the historical number of successful collections of the first sample user of the two sample users is less than the historical number of successful collections of the second sample user, a negative sample is generated; wherein the negative sample includes the parking behavior characteristics, credit index and collection sensitivity index of the two sample users, and a second sample label indicating the historical number of successful collections of the two sample users.

[0024] Optionally, after counting the historical parking data of each sample user to obtain the parking behavior characteristics of the sample user, and calculating the credit index and collection sensitivity index of the sample user based on the historical parking data of the sample user, the method further includes: grouping the sample users according to the historical arrears amount and the historical number of successful collections in the parking behavior characteristics of the sample users to obtain the sample user group to which the sample users belong; wherein the historical arrears amount of the sample users belonging to the same sample user group belongs to the same amount range, and the historical number of successful collections belongs to the same number range;

[0025] The step of inputting the parking behavior characteristics, credit index and collection sensitivity index of each sample user into the collection prediction model of the initial structure to obtain the collection success rate of each sample user includes: for each sample user group, inputting the parking behavior characteristics, credit index and collection sensitivity index of the group of sample users into the collection prediction model of the initial structure corresponding to the sample user group to obtain the collection success rate of the group of sample users;

[0026] The method adjusts the model parameters of the collection prediction model of the initial structure based on the collection success rate of each sample user and the sample labels indicating the historical number of successful collections of each sample user to obtain a trained collection prediction model, including: adjusting the model parameters of the collection prediction model of the initial structure corresponding to the sample user group based on the collection success rate of the group of sample users and the sample labels indicating the historical number of successful collections of the group of sample users to obtain a trained collection prediction model corresponding to the sample user group.

[0027] An embodiment of the present application also provides an electronic device, comprising: a memory for storing computer programs; a processor for executing the program stored in the memory to implement the collection information determination method described in any of the first aspects above, or the model training method described in any of the second aspects above.

[0028] The embodiment of the present application also provides a parking fee management system, the parking fee management system comprising: a data collection module, a processor and a collection module;

[0029] The data collection module is used to collect historical parking data of each candidate user in a historical time period; wherein the candidate user is a user who has outstanding parking fees in the parking lot;

[0030] The processor is used to execute any one of the methods for determining collection information in the first aspect to obtain a target user for recommended collection;

[0031] The collection module is used to collect payment from the target user.

[0032] An embodiment of the present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the method for determining collection information described in any one of the first aspects above, or the model training method described in any one of the second aspects above is implemented.

[0033] An embodiment of the present application also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute any of the collection information determination methods described in the first aspect above, or any of the model training methods described in the second aspect above.

[0034] Beneficial effects of the embodiments of the present application:

[0035] The technical solution provided by the embodiment of the present application obtains historical parking data of each candidate user who has arrears in the parking lot within a historical time period; for each candidate user, based on the historical parking data of the candidate user, determines the parking feature data of the candidate user; the parking feature data includes: the parking behavior characteristics of the candidate user, and / or, the collection sensitivity index; the parking feature data of each candidate user is processed based on a pre-trained collection prediction model, and based on the output result of the collection prediction model, a target user recommended for collection among each candidate user is obtained.

[0036] Based on the above processing, the collection prediction model can obtain the target users for recommended collection based on the parking behavior characteristics of the candidate users and / or the collection sensitivity index, which can improve the accuracy of determining the target users for collection and thus improve the success rate of collection.

[0037] Of course, implementing any product or method of the present application does not necessarily require achieving all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, 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, and for ordinary technicians in this field, other embodiments can also be obtained based on these drawings.

[0039] Figure 1 A structural diagram of a parking fee management system provided in an embodiment of the present application;

[0040] Figure 2 A flowchart of a first method for determining collection information provided in an embodiment of the present application;

[0041] Figure 3 A schematic diagram of historical parking data provided by an embodiment of the present application;

[0042] Figure 4 A schematic diagram of a parking behavior feature provided in an embodiment of the present application;

[0043] Figure 5 A flowchart of a second method for determining collection information provided in an embodiment of the present application;

[0044] Figure 6 A flowchart of a third method for determining payment information provided in an embodiment of the present application;

[0045] Figure 7 A flowchart of a fourth method for determining collection information provided in an embodiment of the present application;

[0046] Figure 8A flowchart of a fifth method for determining collection information provided in an embodiment of the present application;

[0047] Fig. 9 A flowchart of the first model training method provided in an embodiment of the present application;

[0048] Fig.10 A flowchart of a second model training method provided in an embodiment of the present application;

[0049] Fig.11 A flowchart of a third model training method provided in an embodiment of the present application;

[0050] Fig.12 A structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] 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 only 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 based on the present application belong to the scope of protection of the present application.

[0052] In the related art, it is necessary to manually select the target users for collection from the users who owe payment and send collection SMS messages. However, it is time-consuming and laborious to edit a SMS message for each user who owes payment. If collection is collected too frequently, the SMS or traffic cost will be high, and the user may even be blocked. If the collection is collected manually, it is easy to make mistakes and the collection success rate cannot be guaranteed.

[0053] To solve the above problems, see Figure 1 , Figure 1 A structural diagram of a parking fee management system provided in an embodiment of the present application. The parking fee management system includes: a collection module 101, a processor 102 and a collection module 103. The above-mentioned collection module 101, processor 102 and collection module 103 can be deployed on the same physical device. For example, the collection module 101, processor 102 and collection module 103 are deployed on the same server. Alternatively, the above-mentioned collection module 101, processor 102 and collection module 103 can be deployed on different physical devices respectively. For example, the collection module 101 is deployed on a collection device set up in a parking lot. The processor 102 is deployed on a server. The collection module 103 is deployed on a collection terminal.

[0054] A collection device (such as a monitoring device) may be provided in the parking lot, and the collection device may collect parking data of users in the parking lot. The collection module 101 may determine the historical parking data of the candidate user from the parking data collected by the collection device, and send the historical parking data of the candidate user to the processor 102. The processor 102 executes the collection information determination method provided in the embodiment of the present application, processes the historical parking data of the candidate user, obtains the target user recommended for collection, and the processor 102 sends the identifier of the target user to the collection module 103. The collection module 103 collects the target user according to the received identifier of the target user, thereby improving the collection success rate. The collection method for the target user may include sending a collection SMS to the target user, or making a collection call to the user, etc.

[0055] In one implementation, the target user determined is a user with a higher collection success rate, and then the collection is directly made on the target user after the target user is determined.

[0056] In another implementation, after the target user is determined, the collection frequency of the target user is adjusted according to the collection success rate of the target user and the situation of the target user's response to the collection. Then, when the collection frequency corresponding to the adjusted collection frequency is reached, the target user is collected. For example, if the collection success rate of the target user is low and the response to the collection is slow, the collection frequency of the target user is reduced to avoid the problem of high SMS or traffic costs caused by too frequent collection, and reduce the cost of collection, such as adjusting the collection frequency from once a month to once a quarter, and avoiding the situation of being blacklisted by the user due to too frequent collection. If the collection success rate of the target user is high and the response to the collection is fast, the collection frequency of the target user is increased to improve the collection success rate, such as adjusting the collection frequency from once a month to once every two weeks.

[0057] In another implementation, after the target user is determined, the time for urging the target user to pay this time is adjusted based on the time when the target user paid the fee after the last time the target user was urged to pay the fee.

[0058] See also Figure 2 , Figure 2 A flowchart of a method for determining collection information provided in an embodiment of the present application, the method may include the following steps:

[0059] S201: Obtain historical parking data of each candidate user within a historical time period.

[0060] Among them, the candidate users are users who have outstanding fees in the parking lot.

[0061] S202: For each candidate user, determine parking feature data of the candidate user based on the historical parking data of the candidate user.

[0062] The parking characteristic data includes: the parking behavior characteristics of the candidate user, and / or the collection sensitivity index.

[0063] S203: Processing the parking feature data of each candidate user based on the pre-trained collection prediction model, and obtaining the target user recommended for collection among the candidate users based on the output result of the collection prediction model.

[0064] Based on the collection information determination method provided in the embodiment of the present application, the collection prediction model can obtain the recommended target users for collection based on the parking behavior characteristics of the candidate users and / or the collection sensitivity index, which can improve the accuracy of determining the target users for collection and thereby improve the success rate of collection.

[0065] For step S201, the candidate user is a user who has outstanding fees in the parking lot. The electronic device obtains parking data of the candidate user's parking behavior in the parking lot during the historical time period, and performs data cleaning (such as filtering invalid parking data therein) to obtain historical parking data of the candidate user in the parking lot during the historical time period. Invalid parking data refers to data with missing records of parking behavior, for example, only records of vehicle entry but no records of vehicle exit.

[0066] The historical time period can be set according to the needs. For example, the historical time period can be 1 month before the current collection call, or the historical time period can be 3 months before the current collection call, etc.

[0067] In some embodiments, historical parking data includes at least one of the following: vehicle entry data, vehicle exit data, payment data and collection data of the candidate user; vehicle entry data includes: parking lot identification and type; vehicle exit data includes: parking duration and order fee; payment data includes: payment method and payment type; collection data includes: collection amount and collection amount.

[0068] For example, see Figure 3 The parking behaviors of the candidate users include: entering, leaving, paying, and passively urging. Based on the parking behaviors of the candidate users, the historical parking data obtained include: the vehicle entry data, vehicle exit data, payment data, and urging data of the candidate users.

[0069] The vehicle entry data of the candidate user includes: the identification and type of the parking lot for each parking; the types of parking lots include: factory warehouse and roadside.

[0070] The vehicle exit data of the candidate user includes: the parking time and order amount (i.e., order fee) of each parking. The parking time refers to the duration of each parking behavior, that is, the time from the time the vehicle enters the parking lot to the time it leaves the parking lot, and the order amount refers to the parking fee required to be paid for each parking.

[0071] The payment data of the candidate user includes: the payment type and payment method of each payment; the payment types include payment before leaving the parking lot, same-day supplementary payment and cross-day supplementary payment. Payment before leaving the parking lot means payment before the vehicle leaves the parking lot; same-day supplementary payment means payment is made on the same day of parking after the vehicle leaves the parking lot and exceeds the preset payable time. Cross-day supplementary payment means payment is made on a day after the vehicle leaves the parking lot and exceeds the preset payable time.

[0072] The collection data of the candidate user includes: the total collection amount in the historical period, and the total collection amount.

[0073] The payment data of the candidate user includes: the payment method of each payment behavior. The payment methods include: mini program, life account (such as Alipay life account), public account (such as WeChat public account), APP (application), contactless deduction (that is, automatic deduction of parking fees with user authorization), on-site QR code, exit QR code, factory warehouse upload, POS (Point Of Sale) entry ticket QR code (that is, the QR code on the printed receipt), POS upload charging, POS and others.

[0074] With respect to step S202, the parking characteristic data of the candidate user includes: the parking behavior characteristics, the credit index, and at least one of the collection sensitivity index of the candidate user.

[0075] The credit index is used to reflect the number and amount of outstanding orders of the candidate user, as well as the timeliness of the candidate user's voluntary payment. The collection sensitivity index is used to reflect the response speed of the candidate user to make supplementary payments after receiving collection information. The parking behavior characteristics are used to reflect the characteristics of the candidate user's parking, payment and response to collection in the historical time period. The credit index, collection sensitivity index, and parking behavior characteristics can all reflect the user's response to collection. Therefore, based on the candidate user's historical parking data, the candidate user's parking feature data is calculated.

[0076] In some embodiments, the parking behavior characteristics include: statistical characteristics of historical behavior, statistical characteristics of parking behavior, and statistical characteristics of payment behavior.

[0077] For example, see Figure 4, the credit characteristic coefficient is the credit index; the sensitivity of collection is the collection sensitivity index. The statistical characteristics of historical behavior include: the statistical characteristics of historical behavior statistics within the historical time period, and the statistical characteristics of historical behavior in each sub-time period within the historical time period. For example, if the historical time period is 3 months before the current collection, then each month within the historical time period can be set as a sub-time period. Or, if the historical time period is 1 month before the current collection, then each week within the historical time period can be set as a sub-time period.

[0078] The statistical features of the historical behaviors at all times before the current collection (i.e., the historical time period) include: behavior sequence length, number of parking times, number of payment times, number of collection times, number of successful collection times (i.e., the number of successful collection times in the subsequent embodiments), and active days. The behavior sequence length refers to: the total number of parking behaviors in the historical time period. The active days refer to the number of days in the historical time period that the candidate user has parking behaviors.

[0079] The third month before the collection, the second month before the collection, and the first month before the collection are all sub-periods within the historical time period. The statistical characteristics of the historical behavior of the sub-periods include: behavior sequence length, number of parking times, number of payments, number of collections, number of successful collections, and number of active days.

[0080] The statistical characteristics of parking behavior include: the statistical characteristics of parking amount statistics, the statistical characteristics of arrears duration statistics, the total parking fee, the total parking arrears (i.e., the historical arrears amount in the subsequent embodiments), and the most recent parking duration. The statistical characteristics of the parking amount (i.e., parking fee) include: the maximum, minimum, mean, median, standard deviation, and variance of the order fees for each parking. The statistical characteristics of the arrears duration (i.e., the duration from the time when the fee should be paid to the actual time of payment) include: the maximum, minimum, mean, median, standard deviation, and variance of the arrears duration of each parking.

[0081] The statistical characteristics of payment behavior include: the statistical characteristics of payment time (i.e. the time from the departure time to the actual payment time), payment preference and total payment amount. The statistical characteristics of payment time include: the maximum, minimum, mean, median, standard deviation and variance of the payment time of each parking session.

[0082] The payment preference includes: top 1, top 2 and top 3 payment sources of each parking behavior. The payment source is the payment method in the above embodiment.

[0083] In some embodiments, the parking characteristic data of the candidate user includes: the parking behavior characteristics, credit index, and collection sensitivity index of the candidate user. Accordingly, the electronic device can process the historical parking data of the candidate user in the following manner to obtain the parking characteristic data of the candidate user: for each candidate user, the historical parking data of the candidate user is counted to obtain the parking behavior characteristics of the candidate user. Based on the departure time, payment time, and actual payment time of each parking behavior of the candidate user, the payment extension index of the candidate user is calculated, and based on the payment extension index of the candidate user, the credit index of the candidate user is calculated. When the candidate user is not collected, the collection sensitivity index of the candidate user is determined to be a first value; when the number of successful collections in the history of the candidate user is 0, the collection sensitivity index of the candidate user is determined to be a second value; when the number of successful collections in the history of the candidate user is greater than 0, the collection sensitivity index of the candidate user is calculated based on the actual payment time, collection time, and preset payment time of each parking behavior of the candidate user; wherein the preset payment time refers to the time between the departure time of the vehicle and the payment time.

[0084] After obtaining the historical parking data of the candidate user, the historical parking data is counted to obtain the parking behavior characteristics of the candidate user. For example, the payment method of each payment in the historical time period is counted to obtain the payment preference of the candidate user; the success of each collection in the historical time period is counted to obtain the number of successful collections in the history of the candidate user.

[0085] For each parking behavior of the candidate user, the electronic device calculates the payment deferral index of the parking behavior according to the following formula (1) based on the departure time, payment time and actual payment time of the parking behavior.

[0086] (1);

[0087] Indicates the payment deferral index of the candidate user's parking behavior. Indicates the actual payment time for the parking behavior. Indicates the time when the parking fee should be paid; the time when the fee should be paid refers to the time between the departure time of the parking behavior and the preset payment time. Indicates the departure time of this parking behavior.

[0088] It means that the actual payment time of the parking behavior is earlier than the payment time of the parking behavior, that is, the user has paid on time, then the payment deferral index of the candidate user's parking behavior is recorded as 1. It means that the actual payment time of the parking behavior is later than the payment time of the parking behavior, that is, the user has delayed the payment. Less than , then in , the calculated payment deferral index for the candidate user's parking behavior is a value greater than 1.

[0089] Furthermore, based on the payment deferral index of each parking behavior of the candidate user and the following formula (2), the credit index of the candidate user is calculated.

[0090] (2);

[0091] Indicates the credit index of the candidate user; Indicates the total number of parking behaviors of the candidate user; Indicates the number of parking behaviors of the candidate user with delayed payment; represents the payment delay index of the candidate user's i-th parking behavior; represents the amount of parking fees payable by the candidate user for the i-th parking behavior; Indicates the amount of arrears of the candidate user at the time of the i-th parking behavior.

[0092] The electronic device may determine the collection sensitivity index of the candidate user according to the following formula (3) based on the collection situation of the candidate user.

[0093] (3);

[0094] Indicates the collection sensitivity index of the candidate user; Indicates the total number of collection requests for the candidate user; Indicates the number of successful historical collection requests for the candidate user; Indicates the total number of collection requests for the candidate user; Indicates the preset payment period; represents the actual payment time of the kth parking reminder; The payment time is the time when the user receives the payment reminder (such as a payment reminder text message, payment reminder phone call, etc.).

[0095] In the above formula (3), If the value is 0, it means that the candidate user has not been urged to pay, and the payment sensitivity index of the candidate user has no reference significance, and the first value (ie -1) is used to mark the payment sensitivity index of the candidate user as invalid data. A value of 0 indicates that the collection demand for the candidate user has never been successful, and the collection demand sensitivity index of the candidate user is relatively low, so the collection demand sensitivity index of the candidate user is determined to be a second value (ie, 0). If it is greater than 0, it means that the collection of payment for the candidate user has been successful. Based on the actual payment time, collection time and preset payment duration of each parking behavior of the candidate user, the collection sensitivity index of the candidate user is calculated, and the calculated collection sensitivity index is a value greater than 0.

[0096] Based on the above processing, while mining the statistical characteristics of the user's parking behavior, the collection sensitivity index and credit index were used. Subsequently, through the collection prediction model based on machine learning, the collection sensitivity index and credit index obtained by logical calculation and the statistical characteristics of parking behavior were further mined to improve the accuracy of the collection success rate output by the collection prediction model, thereby further improving the collection success rate.

[0097] For step S203, the collection prediction model may be an XGBoost (eXtreme Gradient Boosting) model. Alternatively, it may be a GBDT (Gradient Boosting Decision Tree) model, etc. For each candidate user, the electronic device generates a feature matrix containing the parking feature data of the candidate user, and inputs the feature matrix of each candidate user into the collection prediction model. The collection prediction model includes multiple decision trees, each of which processes the feature matrix of each candidate user to obtain a decision result. The objective function of the collection prediction model processes the decision results of each decision tree to obtain the output result of the collection prediction model.

[0098] The input data of the collection prediction model can be the parking behavior characteristics of the candidate user, or the collection sensitivity index of the candidate user, or the parking behavior characteristics and collection sensitivity index of the candidate user, or the parking behavior characteristics, collection sensitivity index and other parameters (such as credit index) of the candidate user. The output result of the collection prediction model can be the collection success rate of the candidate user, or a label indicating whether to collect the candidate user.

[0099] Accordingly, based on the input data of the collection prediction model and different output results, the electronic device can use different methods to determine the target users for recommended collection in combination with the output results of the collection prediction model.

[0100] Method 1:

[0101] The input data of the collection prediction model is at least one of the parking behavior characteristics and collection sensitivity index of the candidate user. The output result of the collection prediction model is the collection success rate of the candidate user. Accordingly, the electronic device generates a feature matrix of the candidate user, and the feature matrix contains at least one of the parking behavior characteristics and collection sensitivity index of the candidate user. The electronic device inputs the feature matrix of each candidate user into the collection prediction model to obtain the collection success rate of the candidate user finally output by the collection prediction model. After obtaining the collection success rate of each candidate user, the electronic device determines the first third number of candidate users from each candidate user as the target users according to the target sorting result of the collection success rate of each candidate user from high to low.

[0102] Method 2:

[0103] The input data of the collection prediction model are the parking behavior characteristics, collection sensitivity index and credit index of the candidate user. The output result of the collection prediction model is the collection success rate of the candidate user. The collection prediction model is used to predict the collection success rate of each candidate user. The collection success rate can represent the probability of successful collection for each candidate user. The collection success rate output by the collection prediction model is used to rank the candidate users.

[0104] Accordingly, in Figure 2 Based on Figure 5 , step S202 may include the following steps:

[0105] S2021: For each candidate user, based on the candidate user's historical parking data, determine the candidate user's parking behavior characteristics, credit index and collection sensitivity index.

[0106] Accordingly, step S203 may include the following steps:

[0107] S2031: Input the parking behavior characteristics, credit index and collection sensitivity index of each candidate user into a pre-trained collection prediction model to obtain the collection success rate of each candidate user output by the collection prediction model.

[0108] The collection prediction model is obtained by training based on the parking behavior characteristics, credit index and collection sensitivity index of each sample user, as well as sample labels representing the number of historical collection successes of each sample user.

[0109] S2032: Based on the collection success rate of each candidate user, determine the target user recommended for collection from the candidate users.

[0110] In one implementation, the electronic device generates a feature matrix including the parking behavior characteristics, credit index, and collection sensitivity index of the candidate users, and inputs the feature matrix of each candidate user into the collection prediction model. Each decision tree in the collection prediction model processes the feature matrix of each candidate user to obtain a decision result, which is the initial collection success rate of the candidate user. The objective function of the collection prediction model processes the decision results of each decision tree (i.e., the initial collection success rate of the candidate user) to obtain the collection success rate of the candidate user finally output by the collection prediction model. After obtaining the collection success rate of each candidate user, the electronic device determines the first third number of candidate users from each candidate user as the target user according to the target sorting result of the collection success rate of each candidate user from high to low, that is, the target user is a user with a higher collection success rate, which can improve the accuracy of the determined target user and thus improve the collection success rate.

[0111] The third number can be set according to requirements. For example, the third number is 50% of the total number of candidate users, or the third number is 40% of the total number of candidate users.

[0112] In another implementation, Figure 5 Based on Figure 6 After step S2021, the method may further include the following steps:

[0113] S204: Grouping the candidate users based on the parking behavior characteristics of the candidate users to obtain a plurality of candidate user groups.

[0114] Accordingly, step S2031 may include the following steps:

[0115] S20311: For each candidate user group, the parking behavior characteristics, credit index and collection sensitivity index of the group of candidate users are input into the collection prediction model corresponding to the candidate user group to obtain the collection success rate of the group of candidate users.

[0116] Among them, the collection prediction model corresponding to the candidate user group is obtained by training based on the parking behavior characteristics, credit index and collection sensitivity index of each sample user in the corresponding sample user group, and the sample labels representing the historical number of successful collections of each sample user in the sample user group.

[0117] Accordingly, step S2032 may include the following steps:

[0118] S20321: Based on the collection success rate of each group of candidate users, determine the target users recommended for collection from each group of candidate users.

[0119] If there are candidate users with similar parking behavior characteristics among the multiple candidate users, the electronic device can divide the candidate users with similar parking behavior characteristics into a group. Each group of candidate users corresponds to a collection prediction model, which is trained based on the sample user group corresponding to the candidate user group.

[0120] The electronic device may group the candidate users in the following manner:

[0121] Method 1: The parking behavior characteristics of the candidate user include: the historical arrears amount of the candidate user. The electronic device divides the candidate users into multiple candidate user groups according to the historical arrears amount. After the division, the historical arrears amount of the candidate users belonging to the same candidate user group belongs to the same amount range. Different candidate user groups correspond to different amount ranges. In addition, the historical arrears amount of the candidate user group and the corresponding sample user group belongs to the same amount range.

[0122] For example, see Table 1, which is a candidate user grouping table provided in an embodiment of the present application.

[0123] Table 1

[0124] Historical arrears amount Group （0，25] 1 （25，50] 2 （50，10000] 3

[0125] In the grouping method shown in Table 1 above, the candidate users are divided into 3 groups. The historical arrears amount of the candidate users belonging to the first group belongs to the amount range (0, 25). The historical arrears amount of the candidate users belonging to the second group belongs to the amount range (25, 50). The historical arrears amount of the candidate users belonging to the third group belongs to the amount range (50, 10000].

[0126] Method 2: The parking behavior characteristics of the candidate user include: the number of successful collection requests in history of the candidate user. The electronic device divides the candidate user into multiple candidate user groups according to the number of successful collection requests in history. After the division, the number of successful collection requests in history of the candidate users belonging to the same candidate user group belongs to the same number interval. Different candidate user groups correspond to different number intervals. In addition, the number of successful collection requests in history of the candidate user group and the corresponding sample user group belongs to the same number interval.

[0127] For example, see Table 2, which is a candidate user grouping table provided in an embodiment of the present application.

[0128] Table 2

[0129] Historical number of successful collection requests Group （0，5] 1 （5，10] 2 （10，20] 3

[0130] In the grouping method shown in Table 2 above, the candidate users are divided into 3 groups. The number of successful collections in history for the candidate users belonging to the first group belongs to the number interval (0, 5). The number of successful collections in history for the candidate users belonging to the second group belongs to the number interval (5, 10). The number of successful collections in history for the candidate users belonging to the third group belongs to the number interval (10, 20].

[0131] Method 3: The parking behavior characteristics of the candidate user include: the candidate user's historical arrears amount and the historical number of successful collections. Accordingly, the above step S204 may include the following steps: classify the candidate users whose historical arrears amount belongs to the same amount range and whose historical number of successful collections belongs to the same number range into the same candidate user group. Among them, different candidate user groups correspond to different amount ranges and number ranges; the candidate user group and the corresponding sample user group have historical arrears amount in the same amount range, and the historical number of successful collections belongs to the same number range.

[0132] The electronic device divides the candidate users into multiple candidate user groups according to the historical arrears amount and the historical number of successful collection requests. After the division, the historical arrears amount of the candidate users belonging to the same candidate user group belongs to the same amount range, and the historical number of successful collection requests belongs to the same number range; different candidate user groups correspond to different amount ranges and number ranges. In addition, the historical arrears amount of the candidate user group and the corresponding sample user group belongs to the same amount range, and the historical number of successful collection requests belongs to the same number range.

[0133] For example, see Table 3, which is a candidate user grouping table provided in an embodiment of the present application.

[0134] Table 3

[0135]

[0136] In the grouping method shown in Table 3 above, the candidate users are divided into 6 groups. The candidate users belonging to the first group have a historical collection success number of 0, that is, no collection has been successful (No_success), and the historical arrears amount belongs to the amount range (0, 25]. The candidate users belonging to the second group have a historical collection success number of 0, and the historical arrears amount belongs to the amount range (25, 50]. The candidate users belonging to the third group have a historical collection success number of 0, and the historical arrears amount belongs to the amount range (50, 10000]. The candidate users belonging to the fourth group have a historical collection success number greater than 0, that is, the collection has been successful (Success), and the historical arrears amount belongs to the amount range (0, 25]. The candidate users belonging to the fifth group have a historical collection success number greater than 0, and the historical arrears amount belongs to the amount range (25, 50]. The candidate users belonging to the sixth group have a historical collection success number greater than 0, and the historical arrears amount belongs to the amount range (50, 10000].

[0137] After obtaining multiple candidate user groups, for each candidate user group, the electronic device inputs the parking behavior characteristics, credit index and collection sensitivity index of this group of candidate users into the collection prediction model corresponding to the candidate user group, and can obtain the collection success rate of this group of candidate users.

[0138] Based on the above processing, the collection prediction model corresponding to each group of candidate users is trained based on the sample user grouping corresponding to the candidate user grouping. Using the collection prediction model corresponding to each group of candidate users, the collection success rate of the group of candidate users is determined with high accuracy, which further improves the accuracy of the determined target users and improves the collection success rate.

[0139] After obtaining the collection success rate of each group of candidate users, the electronic device determines the recommended collection users from each group of candidate users in the following manner, obtains the recommended collection users in each candidate user group, and obtains the recommended collection target users.

[0140] In some embodiments, step S20321 may include the following steps: for each candidate user group, based on the number of candidate users in the group and the prediction accuracy of the collection prediction model corresponding to the candidate user group for different numbers of users, determine the number of recommended collection users in the group of candidate users as the first number. Wherein, the prediction accuracy is: based on the collection prediction model corresponding to the candidate user group, the collection success rate of the corresponding test user group is predicted and determined. According to the target sorting result of the collection success rate of the group of candidate users from high to low, the first first number of candidate users are determined from the group of candidate users to obtain the recommended collection users in the candidate user group. The recommended collection users in each candidate user group are obtained to obtain the target users for recommended collection.

[0141] For each candidate user group, the electronic device obtains the prediction accuracy of the collection prediction model corresponding to the candidate user group. The prediction accuracy of the collection prediction model corresponding to the candidate user group is obtained by predicting the collection success rate of the test user group corresponding to the candidate user group using the collection prediction model corresponding to the candidate user group. The historical arrears amount of the candidate user group and the corresponding test user group belong to the same amount range, and the historical collection success times belong to the same number range, and the candidate user group and the corresponding test user group correspond to the same collection prediction model.

[0142] The method for determining the prediction accuracy of the collection prediction model corresponding to the candidate user group for different numbers of users includes: for each test user group, the parking behavior characteristics, credit index and collection sensitivity index of the group of test users are input into the corresponding collection prediction model to obtain the collection success rate of the group of test users. The collection success rate of the group of test users is sorted from high to low to obtain the first test sorting result of the group of test users. The group of test users is sorted from high to low according to the number of historical collection successes as the second test sorting result of the group of test users. The first second number of test users in the second test sorting result are compared with the first second number of test users in the first test sorting result to determine the prediction accuracy of the collection prediction model corresponding to the candidate user group corresponding to the test user group for the first second number of users.

[0143] After the collection prediction model corresponding to the candidate user group is trained, the collection prediction model corresponding to the candidate user group is used to process the parking behavior characteristics, credit index and collection sensitivity index of the test user group corresponding to the candidate user group to obtain the collection success rate of this group of test users, and the collection success rates of this group of test users are ranked from high to low to obtain the first test ranking result of this group of test users.

[0144] The test users are sorted from high to low according to the number of successful payment collection in history, as the second test sorting result of the test users. The second test sorting result is the sorting result of the success rate of payment collection of the test users marked in advance.

[0145] Then, determine the top K (i.e., the second largest number) of test users in the first test sorting result whose collection success rates of the test users are from high to low, and compare the top K test users in the first test sorting result with the top K test users in the second test sorting result to obtain the predicted success rate of the collection prediction model corresponding to the test user group for the top K test users. The value of K is set by the technician according to actual needs.

[0146] For example, a group includes 100 test users. In the first test sorting result from high to low collection success rate, the top 10 (i.e., top 10%) test users are determined, and the top 10 test users in the first test sorting result are compared with the top 10 test users in the second test sorting result to obtain the prediction accuracy of the collection prediction model corresponding to the test user group for the top 10 test users. Similarly, the top 50 (i.e., top 50%) test users can also be determined in the first test sorting result from high to low collection success rate, and the top 50 test users in the first test sorting result are compared with the top 50 test users in the second test sorting result to obtain the prediction accuracy of the collection prediction model corresponding to the test user group for the top 50 test users.

[0147] Furthermore, the electronic device obtains the number of users (i.e., the second number) corresponding to the preset threshold value of the prediction accuracy of the collection prediction model, and calculates the ratio of the second number to the total number of the group of test users. Then, the product of the ratio and the number of the group of candidate users is calculated to obtain the number that makes the prediction accuracy of the collection prediction model reach the preset threshold value, i.e., the number of recommended collection users in the group of candidate users, as the first number. The preset threshold value can be set according to demand.

[0148] For example, for the above example, the collection prediction model corresponding to the candidate user group has a prediction accuracy of 80% for the Top 10 test users (i.e., the prediction accuracy for the top 10% of test users is 80%), and the collection prediction model corresponding to the candidate user group has a prediction accuracy of 50% for the Top 50 test users (i.e., the prediction accuracy for the top 50% of test users is 50%).

[0149] If the group contains 100 candidate users. When the preset threshold is 80%, the first number of recommended collection users in the group of candidate users is the top 10% (i.e., 10). When the preset threshold is 50%, the first number of recommended collection users in the group of candidate users is the top 50% (i.e., 50). If the group contains 200 candidate users. When the preset threshold is 80%, the first number of recommended collection users in the group of candidate users is the top 10% (i.e., 20). When the preset threshold is 50%, the first number of recommended collection users in the group of candidate users is the top 50% (i.e., 100).

[0150] Furthermore, for each candidate user group, the electronic device determines the first number of candidate users from the group of candidate users according to the target sorting result from high to low according to the collection success rate of the group of candidate users, obtains the recommended collection users in the candidate user group, and obtains the recommended collection users in each candidate user group to obtain the target users for recommended collection.

[0151] Based on the above processing, the target ranking results of the collection success rate of each group of candidate users are determined respectively, and the recommended collection users of the group of candidate users are determined according to the target ranking results of the collection success rate of each group of candidate users, which can improve the accuracy of the recommended collection users, further improve the accuracy of the target users, and improve the collection success rate. That is, by adopting the strategy of multi-dimensional grouping and dynamic weighting, the collection success rate of candidate users is first sorted according to the candidate user group, and the dynamic ranking list of each group (i.e., recommended collection users) is obtained by combining the accuracy of the collection prediction modules of different groups, and finally the lists of each group are combined to obtain the final recommended ranking list (i.e., target users), which greatly improves the collection success rate and the collection amount.

[0152] In yet another implementation, the target user may be determined based on the collection success rate of the candidate users and the response of the candidate users to the collection.

[0153] The candidate user's response to the collection call reflects the candidate user's response to the collection call, the payment extension, and the timeliness of the voluntary payment. The candidate user's response to the collection call can be represented by a target value. The target value is determined based on at least one of the candidate user's payment extension index, credit index, and collection sensitivity index. The larger the target value, the more positive the candidate user's response to the collection call.

[0154] Exemplarily, the target value is determined based on the payment deferral index of the candidate user. For example, the statistical value (e.g., mean, variance, maximum value, etc.) of the payment deferral index of each parking behavior of the candidate user is calculated, and the statistical value is normalized to the interval [0, 1], and then the difference between the third value and the normalized statistical value is calculated as the target value. The third value is 1. Alternatively, one of the credit index or the collection sensitivity index of the candidate user is selected as the target value.

[0155] Correspondingly, the electronic device determines a target user for recommended payment collection from among the candidate users based on the payment collection success rate and the target value of each candidate user.

[0156] Exemplarily, without grouping the candidate users, the electronic device calculates the weighted sum of the collection success rate and the target value of the candidate users, and determines the first third number of candidate users as target users in descending order of the weighted results. Alternatively, the electronic device sorts the candidate users in descending order according to the collection success rate of the candidate users, and selects the first fourth number of candidate users (which may be referred to as to-be-processed users) from the sorted candidate users, where the fourth number is greater than the third number. Then, in descending order of the target values, the first third number of to-be-processed users are determined to be the target users recommended for collection.

[0157] When the candidate users are grouped, for each candidate user group, the electronic device first determines the number of recommended payment collection users in the candidate user group (i.e., the first number), and the method of determining the first number is referred to the relevant introduction of the above embodiment. Then, based on the collection success rate and target value of the group of candidate users, the recommended payment collection users in the group of candidate users are determined, and the recommended payment collection users in each group of candidate users are obtained to obtain the target users for recommended payment collection.

[0158] Exemplarily, for each candidate user group, the electronic device calculates the weighted sum of the collection success rate and the target value of the group of candidate users, and determines the first number of candidate users in descending order of the weighted results to obtain the recommended collection users in the group of candidate users. Alternatively, for each candidate user group, the electronic device sorts the candidate users in descending order according to the collection success rate of the group of candidate users, and selects the first fourth number of candidate users (which may be referred to as to-be-processed users) from the sorted group of candidate users, where the fourth number is greater than the first number. Then, the first number of to-be-processed users is determined in descending order of the target values ​​to obtain the recommended collection users in the group of candidate users.

[0159] Based on the above processing, the target users are determined based on the collection success rate of the candidate users and their responses to the collection. This not only takes into account the collection success of the candidate users, but also takes into account the response of the candidate users to the collection and the extension of payment, etc. This can further improve the accuracy of the determined target users and the collection success rate.

[0160] Method three:

[0161] The input data of the collection prediction model is at least one of the parking behavior characteristics of the candidate user and the collection sensitivity index. The output result of the collection prediction model is a label indicating whether to collect the payment from the candidate user.

[0162] Accordingly, in Figure 2 Based on Figure 7 , step S203 may include the following steps:

[0163] S2033: Input the parking feature data of each candidate user into a pre-trained collection prediction model to obtain the target user for collection recommended by the collection prediction model.

[0164] The collection prediction model is obtained by training based on the parking feature data of each sample user and a sample label indicating whether the collection of each sample user is successful.

[0165] The electronic device generates a feature matrix of the candidate user, which includes at least one of the candidate user's parking behavior features and a collection sensitivity index. The electronic device inputs the feature matrix of each candidate user into the collection prediction model, and obtains a label output by the collection prediction model indicating whether to collect the candidate user. For example, a label of 1 indicates that the candidate user is collected, and a label of 0 indicates that the candidate user is not collected.

[0166] In one implementation, when the candidate users are not grouped, the electronic device determines, from the candidate users, a candidate user whose tag indicates that the candidate user is urged to pay (ie, a candidate user whose tag is 1) as the target user.

[0167] In another implementation, when the candidate users are grouped, for each candidate user group, the electronic device first determines the number of recommended payment collection users (i.e., the first number) in the candidate user group. The method for determining the first number is described in the above-mentioned embodiment. Then, for each candidate user group, the first number of candidate users whose tags indicate that the candidate users are to be collected are determined from the group of candidate users, and the recommended payment collection users in the group of candidate users are obtained. Then, the recommended payment collection users in each group of candidate users are obtained to obtain the target users.

[0168] Based on the above processing, the collection prediction model directly outputs a label of whether to collect payment from the candidate user. The target user can be determined directly based on the output result of the collection prediction model, thereby improving the efficiency of determining the target user and further improving the collection efficiency.

[0169] In another implementation, the target user may be determined by combining a label indicating whether to urge the candidate user to pay and a target value reflecting the candidate user's response to the urging. The method for determining the candidate user's response to the urging may refer to the relevant introduction of the above embodiment.

[0170] Exemplarily, without grouping the candidate users, the electronic device determines from the candidate users the candidate users whose tags indicate that they are urging payment as the users to be processed, and selects the first third number of the users to be processed in descending order of the target values ​​to obtain the target user.

[0171] Alternatively, when the candidate users are grouped, for each candidate user group, the electronic device first determines the number of recommended payment collection users (i.e., the first number) in the candidate user group. The method for determining the first number is described in the related description of the above-mentioned embodiment. Then, for each candidate user group, the candidate users in the group of candidate users whose tags indicate payment collection for the candidate users are determined as the users to be processed, and the first first number of users to be processed are selected in descending order of the target values ​​of the group of candidate users to obtain the recommended payment collection users in the group of candidate users, and then the recommended payment collection users in each group of candidate users are obtained to obtain the target users.

[0172] Based on the above processing, the target users are determined based on the output results of the collection prediction model and the response of the candidate users to the collection. This not only takes into account the success of the collection of the candidate users, but also takes into account the response of the candidate users to the collection and the extension of payment, etc. This can further improve the accuracy of the determined target users and increase the success rate of collection.

[0173] In some embodiments, Figure 5 Based on Figure 8 After step S2032, the method may further include the following steps:

[0174] S205: Target users whose collection success rates belong to the same success rate range are divided into a target user group to obtain a plurality of target user groups.

[0175] S206: For each target user group, based on the success rate interval to which the collection success rate of the target users in the group belongs and the corresponding relationship between the preset success rate interval and the collection frequency, determine the collection frequency of the target users in the group.

[0176] S207: urging the target users in the group to pay according to their collection frequency.

[0177] In one implementation, after determining the target users, the electronic device may divide the target users whose collection success rates belong to the same success rate range into a target user group, thereby obtaining a plurality of target user groups.

[0178] For example, see Table 4, which is a target user grouping table provided in an embodiment of the present application.

[0179] Table 4

[0180] Collection success rate Group （0，a1] 1 (a1, a2] 2 (a2, a3] 3

[0181] In the grouping method shown in Table 4 above, the target users are divided into 3 groups. The collection success rate of the first group of target users belongs to the success rate interval (0, a1]. The collection success rate of the second group of target users belongs to the success rate interval (a1, a2]. The collection success rate of the third group of target users belongs to the success rate interval (a2, a3]. Among them, a1 is less than a2, and a2 is less than a3.

[0182] In the preset correspondence between the success rate interval and the collection frequency, the larger the upper limit of the success rate interval is, the higher the corresponding collection frequency is. Then, for each target user group, in the preset correspondence between the success rate interval and the collection frequency, the collection frequency corresponding to the success rate interval to which the collection success rate of the target user group belongs is determined, the collection frequency of the target user group is obtained, and the collection time of the target user group is determined according to the collection frequency of the target user group, and when the corresponding collection time is reached, the target user group is collected.

[0183] Exemplarily, for the three target user groups shown in Table 4 above, the collection success rate of the third group of target users is relatively high, and the collection frequency of the third group of target users is relatively high. For example, the third group of target users is collected on the same day. The collection success rate of the second group of target users is lower than that of the third group of target users, and the collection frequency of the second group of target users is lower than that of the third group of target users. For example, the second group of target users is collected one week later. The collection success rate of the first group of target users is lower than that of the second group of target users, and the collection frequency of the first group of target users is lower than that of the second group of target users. For example, the first group of target users is collected two weeks later.

[0184] Alternatively, if the collection success rate of the third group of target users is relatively high, the collection frequency of the third group of target users is relatively high, for example, the collection frequency of the third group of target users is adjusted to once a week. Accordingly, the electronic device determines the time one week after the last collection of the third group of target users as the current collection time, and collects the collection of the third group of target users when the current collection time is reached.

[0185] If the collection success rate of the second group of target users is lower than that of the third group of target users, the collection frequency of the second group of target users is lower than that of the third group of target users. For example, the collection frequency of the second group of target users is adjusted to once every two weeks. Accordingly, the electronic device determines the time two weeks after the last collection of the second group of target users as the current collection time, and collects the collection from the second group of target users when the current collection time is reached.

[0186] If the collection success rate of the first group of target users is lower than that of the second group of target users, the collection frequency of the first group of target users is lower than that of the second group of target users. For example, the collection frequency of the first group of target users is adjusted to once a month. Accordingly, the electronic device determines the time one month after the last collection of the first group of target users as the current collection time, and collects the collection from the first group of target users when the current collection time is reached.

[0187] Based on the above processing, the collection frequency of the target users is determined based on the collection success rate of the target users. The determined collection frequency is more suitable for the corresponding target user grouping. It can not only avoid the situation where the SMS or traffic costs are high due to too frequent collection, thereby reducing the collection costs, but also avoid the situation where the collection frequency is too low, resulting in a low collection success rate, thereby further improving the collection success rate.

[0188] In another implementation, after determining the target users, the electronic device can group the target users based on the collection success rate of the target users and the target value indicating the response of the target users to the collection. For example, target users whose collection success rates belong to the same success rate interval and whose target values ​​belong to the same value interval are grouped into one target user group, thereby obtaining multiple target user groups.

[0189] For example, see Table 5, which is a target user grouping table provided in an embodiment of the present application.

[0190] Table 5

[0191] Collection success rate Target value Group （0，a1] （0，T1] 1 (a1, a2] (T1, T2] 2 (a2, a3] (T2, T3] 3

[0192] In the grouping method shown in Table 5 above, the target users are divided into 3 groups. The collection success rate of the first group of target users belongs to the success rate interval (0, a1], and the target value belongs to the value interval (0, T1]. The collection success rate of the second group of target users belongs to the success rate interval (a1, a2], and the target value belongs to the value interval (T1, T2]. The collection success rate of the third group of target users belongs to the success rate interval (a2, a3], and the target value belongs to the value interval (T2, T3]. Among them, a1 is less than a2, a2 ​​is less than a3. T1 is less than T2, and T2 is less than T3.

[0193] The larger the upper limit of the success rate interval and the larger the upper limit of the numerical interval, the higher the corresponding collection frequency. Then, for each target user group, in the preset correspondence between the success rate interval, the numerical interval and the collection frequency, the success rate interval to which the collection success rate of the target user group belongs and the collection frequency corresponding to the numerical interval to which the target numerical value of the target user group belongs are determined, and the collection frequency of the target user group is obtained. According to the collection frequency of the target user group, the collection time of the target user group is determined, and when the corresponding collection time is reached, the target user group is collected.

[0194] Exemplarily, for the three target user groups shown in Table 5 above, the collection success rate and target value of the third group of target users are relatively high, and the collection frequency of the third group of target users is relatively high. For example, the third group of target users is collected on the same day. The collection success rate of the second group of target users is lower than the collection success rate of the third group of target users, and the target value of the second group of target users is lower than the target value of the third group of target users, then the collection frequency of the second group of target users is lower than the collection frequency of the third group of target users. For example, the second group of target users is collected one week later. The collection success rate of the first group of target users is lower than the collection success rate of the second group of target users, and the target value of the first group of target users is lower than the target value of the second group of target users, then the collection frequency of the first group of target users is lower than the collection frequency of the second group of target users. For example, the first group of target users is collected two weeks later.

[0195] Alternatively, if the collection success rate and target value of the third group of target users are both relatively high, the collection frequency of the third group of target users is relatively high. For example, the collection frequency of the third group of target users is adjusted to once a week. Accordingly, the electronic device determines the time one week after the last collection of the third group of target users as the current collection time, and collects the collection of the third group of target users when the current collection time is reached.

[0196] If the collection success rate of the second group of target users is lower than that of the third group of target users, and the target value of the second group of target users is lower than that of the third group of target users, then the collection frequency of the second group of target users is lower than that of the third group of target users. For example, the collection frequency of the second group of target users is adjusted to once every two weeks. Accordingly, the electronic device determines that the time two weeks after the last collection of the second group of target users is the current collection time, and collects the collection from the second group of target users when the current collection time is reached.

[0197] If the collection success rate of the first group of target users is lower than that of the second group of target users, and the target value of the first group of target users is lower than that of the second group of target users, then the collection frequency of the first group of target users is lower than that of the second group of target users. For example, the collection frequency of the first group of target users is adjusted to once a month. Accordingly, the electronic device determines the time one month after the last collection of the first group of target users as the current collection time, and collects the collection from the first group of target users when the current collection time is reached.

[0198] Based on the above processing, the collection frequency of the target users is determined based on the collection success rate of the target users and combined with the target users' response to the collection. The determined collection frequency is more suitable for the corresponding target user grouping. It can not only avoid the situation where high SMS or traffic costs are caused by too frequent collection, thereby reducing the collection cost, but also avoid the situation where the collection frequency is too low, resulting in a low collection success rate, thereby further improving the collection success rate.

[0199] In some embodiments, the electronic device can determine the target user recommended for collection during the collection period from the candidate users according to the collection information determination method provided in the embodiments of the present application when the collection period is reached. Accordingly, during different collection periods, the collection frequency of the target user changes dynamically, that is, the collection frequency of different users can be dynamically adjusted, further improving the collection success rate.

[0200] The following is an explanation of the method for training the collection prediction model with examples. It is understandable that the input data of the collection prediction model when training the collection prediction model corresponds to the input data when using the collection prediction model for prediction, and the output result of the collection prediction model when training the collection prediction model corresponds to the output result when using the collection prediction model for prediction.

[0201] For example, if the input data of the collection prediction model when training the collection prediction model is the parking behavior characteristics, credit index and collection sensitivity index of the sample user, then the input data when using the collection prediction model for prediction is the parking behavior characteristics, credit index and collection sensitivity index of the candidate user. If the output result of the collection prediction model when training the collection prediction model is the collection success rate of the sample user, then the output result when using the collection prediction model for prediction is the collection success rate of the candidate user.

[0202] Alternatively, if the input data of the collection prediction model when training the collection prediction model is the parking behavior characteristics and collection sensitivity index of the sample user, then the input data when using the collection prediction model for prediction is the parking behavior characteristics and collection sensitivity index of the candidate user. If the output result of the collection prediction model when training the collection prediction model is a label indicating whether to collect the sample user, then the output result when using the collection prediction model for prediction is a label indicating whether to collect the candidate user.

[0203] Method 1:

[0204] The input data of the collection prediction model is at least one of the parking behavior characteristics of the candidate user and the collection sensitivity index. The output result of the collection prediction model is the collection success rate of the candidate user.

[0205] Accordingly, the electronic device obtains the historical parking data of the sample user, and determines the parking behavior characteristics and the collection sensitivity index of the sample user based on the historical parking data of the sample user. The manner in which the electronic device determines the parking behavior characteristics and the collection sensitivity index of the sample user is similar to the manner in which the parking behavior characteristics and the collection sensitivity index of the candidate user are determined, and reference is made to the relevant description of the aforementioned embodiment.

[0206] Then, the electronic device generates a feature matrix of the sample user, which includes at least one of the parking behavior characteristics and the collection sensitivity index of the sample user. The electronic device inputs the feature matrix of the sample user into the collection prediction model of the initial structure to obtain the collection success rate of the sample user output by the collection prediction model of the initial structure. The number of historical collection successes of the sample user is obtained, and the number of historical collection successes of the sample user is normalized to [0, 1] as the collection success rate of the pre-labeled sample user.

[0207] Then, based on the collection success rate of sample users output by the collection prediction model of the initial structure and the collection success rate of sample users marked in advance, the function value of the preset loss function is calculated, and the calculated function value represents: the difference between the collection success rate of sample users output by the collection prediction model of the initial structure and the collection success rate of sample users marked in advance. Accordingly, based on the calculated function value, the model parameters of the collection prediction model of the initial structure are adjusted, so that the difference between the collection success rate of sample users output by the collection prediction model of the initial structure and the collection success rate of sample users marked in advance gradually decreases until the preset convergence condition is reached, and a trained collection prediction model is obtained.

[0208] Method 2:

[0209] The input data of the collection prediction model are the parking behavior characteristics, collection sensitivity index and credit index of the candidate user. The output result of the collection prediction model is the collection success rate of the candidate user. Fig. 9 , Fig. 9 A flowchart of a model training method provided in an embodiment of the present application, the method may include the following steps:

[0210] S901: Obtain historical parking data of each sample user in a historical time period.

[0211] S902: Generate training samples based on the historical parking data of each sample user.

[0212] The training samples include: parking behavior characteristics, credit index and collection sensitivity index of each sample user, and sample labels indicating the number of successful collections in history of each sample user.

[0213] S903: Input the parking behavior characteristics, credit index and collection sensitivity index of each sample user into the collection prediction model of the initial structure to obtain the collection success rate of each sample user.

[0214] S904: Based on the collection success rate of each sample user and the sample label indicating the number of historical collection successes of each sample user, the model parameters of the collection prediction model of the initial structure are adjusted to obtain a trained collection prediction model.

[0215] Based on the model training method provided in the embodiment of the present application, a collection collection prediction model can be trained. The subsequent collection collection prediction model is used to determine the target user recommended for collection collection from the candidate users, thereby improving the accuracy of collection collection.

[0216] With respect to step S901 and step S902 , the historical data of the sample user in the historical time period is similar to the historical parking data of the candidate user in the historical time period, and reference may be made to the relevant introduction of the aforementioned embodiment.

[0217] In one implementation, the electronic device processes the historical parking data of the sample user according to a (Pairwise Approach) strategy to construct a training sample for training the collection prediction model.

[0218] In some embodiments, the training samples include positive samples and negative samples. Fig. 9 Based on Fig.10 , step S902 may include the following steps:

[0219] S9021: For each sample user, the historical parking data of the sample user is counted to obtain the parking behavior characteristics of the sample user, and based on the historical parking data of the sample user, the credit index and collection sensitivity index of the sample user are calculated.

[0220] S9022: For every two sample users, if the number of successful historical collection requests of the first sample user among the two sample users is greater than the number of successful historical collection requests of the second sample user, a positive sample is generated.

[0221] The positive samples include the parking behavior characteristics, credit index and collection sensitivity index of the two sample users, and the first sample labels representing the number of successful collections in history of the two sample users.

[0222] S9023: If the number of successful collection requests in history for the first sample user among the two sample users is less than the number of successful collection requests in history for the second sample user, a negative sample is generated.

[0223] The negative samples include the parking behavior characteristics, credit index and collection sensitivity index of the two sample users, and a second sample label indicating the number of successful collections in history of the two sample users.

[0224] The way in which the electronic device determines the parking behavior characteristics, credit index and collection sensitivity index of the sample user is similar to the way in which the parking behavior characteristics, credit index and collection sensitivity index of the candidate user are determined, and reference is made to the relevant introduction of the aforementioned embodiment.

[0225] For each sample user, the number of successful collections in history of the sample user is obtained. The electronic device pairs each sample user according to the Pairwise Approach strategy. Specifically, for every two sample users, if the number of successful collections in history of the first sample user of the two sample users is greater than the number of successful collections in history of the second sample user, then it is determined that the two sample users constitute a positive sample, that is, a positive sample is generated containing the parking behavior characteristics, credit index and collection sensitivity index of the two sample users, and a first sample label representing the number of successful collections in history of the two sample users. The first sample label is represented by a value of 1, that is, the first sample label of the positive sample is 1, and the number of successful collections in history of the first sample user in the positive sample is greater than the number of successful collections in history of the second sample user.

[0226] If the number of successful collections in history of the first sample user of the two sample users is less than the number of successful collections in history of the second sample user, the two sample users are determined to form a negative sample, that is, a negative sample is generated that includes the parking behavior characteristics, credit index and collection sensitivity index of the two sample users, and a second sample label indicating the number of successful collections in history of the two sample users. The second sample label is represented by a value of 0, that is, the second sample label of the negative sample is 0, and the number of successful collections in history of the second sample user in the positive sample is less than the number of successful collections in history of the second sample user.

[0227] Based on the above processing, the sample labels are determined using the number of successful historical collection requests, and a collection prediction model based on machine learning is introduced. The collection prediction model is trained through supervised learning to improve the accuracy of the subsequent collection prediction model output results and further improve the collection success rate.

[0228] In another implementation, the electronic device may process the historical parking data of the sample user according to the Pointwise Approach strategy to construct a training sample for training the collection prediction model.

[0229] Specifically, for each sample user, the number of successful collections in history of the sample user is obtained. If the number of successful collections in history of the sample user is greater than 0, that is, the collection has been successful in history, then the sample user is determined to be a positive sample, that is, a positive sample of the first sample label containing the parking behavior characteristics, credit index and collection sensitivity index of the sample user, and the number of successful collections in history of the sample user is generated. The first sample label is represented by a numerical value of 1. If the number of successful collections in history of the sample user is 0, that is, there has been no successful collection in history, then the sample user is determined to be a negative sample, that is, a negative sample of the second sample label containing the parking behavior characteristics, credit index and collection sensitivity index of the sample user, and the number of successful collections in history of the sample user is generated. The second sample label is represented by a numerical value of 0.

[0230] In another implementation, the electronic device may process the historical parking data of the sample user according to the (Listwise Approach) strategy to construct a training sample for training the collection prediction model. Specifically, for each sample user, the number of successful collections in history of the sample user is obtained. The sample users are sorted from high to low according to their successful collections in history to obtain a first user sequence, and a training sample containing the parking behavior characteristics, credit index and collection sensitivity index of each sample user, as well as a sample label representing the first user sequence is generated.

[0231] With respect to step S903 and step S904, when the electronic device constructs training samples according to the Pairwise Approach strategy, the electronic device trains the collection prediction model in the following manner.

[0232] In one implementation, without grouping the sample users, the electronic device inputs the parking behavior characteristics, credit index, and collection sensitivity index of each sample user into the collection prediction model of the initial structure to obtain the collection success rate of each sample user. Based on the collection success rate of each sample user, the electronic device determines a prediction label representing the collection success rate of every two sample users. For every two sample users, if the collection success rate of the first sample user of the two sample users is greater than the collection success rate of the second sample user, the predicted sample label is 1. If the collection success rate of the first sample user of the two sample users is less than the collection success rate of the second sample user, the predicted sample label is 0.

[0233] Furthermore, based on the predicted labels representing the collection success rate of every two sample users and the sample labels of the training samples composed of the two sample users, the function value of the preset loss function is calculated, and the calculated function value represents the difference between the predicted label and the sample label. If the first sample user and the second sample user of the two sample users form a positive sample, the predicted sample label is 1, and if the first sample user and the second sample user of the two sample users form a negative sample, the predicted sample label is 0. Accordingly, based on the calculated function value, the model parameters of the collection prediction model of the initial structure are adjusted so that the difference between the predicted label and the sample label is gradually reduced until the preset convergence condition is reached, and a trained collection prediction model is obtained.

[0234] In another implementation, the electronic device groups sample users to obtain a plurality of sample user groups, and for each sample user group, trains a collection prediction model corresponding to the sample user group.

[0235] In some embodiments, Fig. 9 Based on Fig.11 After step S9021, the method may further include the following steps:

[0236] S905: Grouping the sample users according to the historical arrears amount and the historical number of successful collection requests in the parking behavior characteristics of the sample users to obtain the sample user group to which the sample users belong.

[0237] Among them, the historical arrears amount of sample users belonging to the same sample user group belongs to the same amount range, and the historical number of successful collection requests belongs to the same number range.

[0238] Accordingly, step S903 may include the following steps:

[0239] S9031: For each sample user group, the parking behavior characteristics, credit index and collection sensitivity index of the sample user group are input into the collection prediction model of the initial structure corresponding to the sample user group to obtain the collection success rate of the sample user group.

[0240] Accordingly, step S904 may include the following steps:

[0241] S9041: Based on the collection success rate of the group of sample users and the sample labels indicating the number of historical collection successes of each sample user, the model parameters of the collection prediction model of the initial structure corresponding to the sample user group are adjusted to obtain a trained collection prediction model corresponding to the sample user group.

[0242] The way in which the electronic device groups the sample users is similar to the way in which the electronic device groups the candidate users, and reference may be made to the related introduction of the foregoing embodiment.

[0243] After obtaining multiple sample user groups, for each sample user group, the electronic device inputs the parking behavior characteristics, credit index, and collection sensitivity index of the sample users in the group into the collection prediction model of the initial structure to obtain the collection success rate of the sample users in the group. Furthermore, based on the collection success rate of the sample users in the group, the electronic device determines a prediction label representing the collection success rate of every two sample users in the group. For every two sample users, if the collection success rate of the first sample user of the two sample users is greater than the collection success rate of the second sample user, the predicted sample label is 1. If the collection success rate of the first sample user of the two sample users is less than the collection success rate of the second sample user, the predicted sample label is 0.

[0244] Furthermore, for each sample user group, based on the predicted labels representing the collection success rate of each two sample users in the group and the sample labels of the training samples composed of the two sample users in the group, the function value of the preset loss function is calculated, and the calculated function value represents the difference between the predicted label and the sample label. If the first sample user and the second sample user of the two sample users form a positive sample, the predicted sample label is 1, and if the first sample user and the second sample user of the two sample users form a negative sample, the predicted sample label is 0. Accordingly, based on the calculated function value, the model parameters of the collection prediction model of the initial structure are adjusted so that the difference between the predicted label and the sample label gradually decreases until the preset convergence condition is reached, and the trained collection prediction model corresponding to the sample user group is obtained.

[0245] Based on the above processing, the sample users are divided into multiple groups, and the collection prediction model training is performed on the grouped sample users respectively, so as to obtain the collection prediction model training corresponding to each group of sample users. Subsequently, the collection prediction model corresponding to the sample users corresponding to each group of candidate users is used to determine that the collection success rate of the group of candidate users is more accurate, further improving the accuracy of the determined target users and improving the collection success rate.

[0246] When the electronic device constructs training samples according to the Pointwise Approach strategy, the electronic device trains the collection prediction model in the following manner.

[0247] In one implementation, without grouping the sample users, the electronic device inputs the parking behavior characteristics, credit index, and collection sensitivity index of each sample user into the collection prediction model of the initial structure to obtain the collection success rate of each sample user. Then, based on the collection success rate of each sample user and the sample label representing the number of historical collection successes of each sample user, the function value of the preset loss function is calculated, and the calculated function value represents the difference between the collection success rate and the sample label. Accordingly, based on the calculated function value, the model parameters of the collection prediction model of the initial structure are adjusted so that the difference between the prediction label and the sample label gradually decreases until the preset convergence condition is reached, and a trained collection prediction model is obtained.

[0248] In another implementation, the electronic device groups the sample users to obtain a plurality of sample user groups, and for each sample user group, trains and obtains a collection prediction model corresponding to the sample user group. For each sample user group, the electronic device inputs the parking behavior characteristics, credit index, and collection sensitivity index of the group of sample users into the collection prediction model of the initial structure to obtain the collection success rate of the group of sample users. Furthermore, based on the collection success rate of the group of sample users and the sample labels representing the number of historical collection successes of the group of sample users, the function value of the preset loss function is calculated, and the calculated function value represents the difference between the collection success rate and the sample label. Accordingly, based on the calculated function value, the model parameters of the collection prediction model of the initial structure are adjusted so that the difference between the prediction label and the sample label is gradually reduced until the preset convergence condition is reached, and the trained collection prediction model corresponding to the sample user group is obtained.

[0249] When the electronic device constructs training samples according to the Listwise Approach strategy, the electronic device trains the collection prediction model in the following manner.

[0250] In one implementation, without grouping the sample users, the electronic device inputs the parking behavior characteristics, credit index and collection sensitivity index of each sample user into the collection prediction model of the initial structure to obtain the collection success rate of each sample user. Then, the sample users are sorted from high to low according to the collection success rate to obtain a second user sequence. Based on the sample labels representing the first user sequence and the second user sequence, the function value of the preset loss function is calculated, and the calculated function value represents the difference between the first user sequence and the second user sequence. Accordingly, based on the calculated function value, the model parameters of the collection prediction model of the initial structure are adjusted so that the difference between the first user sequence and the second user sequence gradually decreases until the preset convergence condition is reached, and a trained collection prediction model is obtained.

[0251] In another implementation, the electronic device groups the sample users to obtain a plurality of sample user groups, and for each sample user group, trains and obtains a collection prediction model corresponding to the sample user group. For each sample user group, the electronic device inputs the parking behavior characteristics, credit index and collection sensitivity index of the group of sample users into the collection prediction model of the initial structure to obtain the collection success rate of the group of sample users. Then, the collection success rate of the group of sample users is sorted from high to low to obtain the second user sequence of the group of sample users. Based on the sample labels representing the first user sequence of the group of sample users and the second user sequence, the function value of the preset loss function is calculated, and the calculated function value represents the difference between the first user sequence and the second user sequence of the group of sample users. Accordingly, based on the calculated function value, the model parameters of the collection prediction model of the initial structure are adjusted so that the difference between the first user sequence and the second user sequence of the group of sample users gradually decreases until the preset convergence condition is reached, and the trained collection prediction model corresponding to the sample user group is obtained.

[0252] Method three:

[0253] The input data of the collection prediction model is at least one of the parking behavior characteristics of the candidate user and the collection sensitivity index. The output result of the collection prediction model is a label indicating whether to collect the candidate user.

[0254] Accordingly, the electronic device obtains the historical parking data of the sample user, and determines the parking behavior characteristics and the collection sensitivity index of the sample user based on the historical parking data of the sample user. The manner in which the electronic device determines the parking behavior characteristics and the collection sensitivity index of the sample user is similar to the manner in which the parking behavior characteristics and the collection sensitivity index of the candidate user are determined, and reference is made to the relevant description of the aforementioned embodiment.

[0255] Then, the electronic device generates a feature matrix of the sample user, which includes at least one of the parking behavior characteristics of the sample user and the collection sensitivity index. The electronic device inputs the feature matrix of the sample user into the collection prediction model of the initial structure to obtain a label output by the collection prediction model of the initial structure indicating whether to collect the sample user, for example, a label of 1 indicates that the sample user is collected, and a label of 0 indicates that the sample user is not collected.

[0256] Furthermore, if the sample user has been successfully urged to pay in the past, the label of the pre-marked sample user is determined to be 1; if the sample user has been unsuccessfully urged to pay in the past, the label of the pre-marked sample user is determined to be 0.

[0257] Furthermore, based on the label indicating whether to collect payment from the sample user and the label of the sample user labeled in advance, the function value of the preset loss function is calculated, and the calculated function value indicates: the difference between the label indicating whether to collect payment from the sample user and the label of the sample user labeled in advance, which is output by the collection prediction model of the initial structure. Accordingly, the model parameters of the collection prediction model of the initial structure are adjusted based on the calculated function value, so that the difference between the label indicating whether to collect payment from the sample user and the label of the sample user labeled in advance gradually decreases, until the preset convergence condition is reached, and a trained collection prediction model is obtained.

[0258] In the technical solution of this application, the operations involved in obtaining, storing, using, processing, transmitting, providing and disclosing user personal information are all carried out with the user's authorization.

[0259] The present application also provides an electronic device, such as Fig.12 As shown, it includes: a memory 1201 for storing computer programs; a processor 1202, for implementing the steps of any of the above-mentioned methods for determining collection information, or any of the above-mentioned steps of the model training method when executing the program stored in the memory 1201.

[0260] In addition, the electronic device may further include a communication bus and / or a communication interface, and the processor 1202, the communication interface, and the memory 1201 communicate with each other via the communication bus. The communication bus mentioned in the electronic device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface is used for communication between the electronic device and other devices.

[0261] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0262] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0263] In another embodiment provided in the present application, a computer-readable storage medium is also provided, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods for determining collection information or any of the above-mentioned steps of the model training method are implemented.

[0264] In another embodiment provided in the present application, a computer program product comprising instructions is also provided, which, when executed on a computer, enables the computer to execute any of the collection information determination methods in the above embodiments, or any of the above model training methods.

[0265] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a solid-state hard disk (SSD), etc.

[0266] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0267] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, electronic device, computer-readable storage medium, and computer program product embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0268] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application are included in the protection scope of the present application.

Claims

1. A method for determining collection information, characterized in that: The method comprises: Acquire historical parking data of each candidate user in a historical time period; wherein the candidate user is a user who has outstanding parking fees in the parking lot; For each candidate user, based on the historical parking data of the candidate user, determine the parking characteristic data of the candidate user; wherein the parking characteristic data includes: the parking behavior characteristics of the candidate user, and / or the collection sensitivity index; Processing the parking feature data of each candidate user based on a pre-trained collection prediction model, and obtaining a target user recommended for collection among each candidate user based on an output result of the collection prediction model; The step of determining, for each candidate user, parking feature data of the candidate user based on the historical parking data of the candidate user includes: For each candidate user, based on the candidate user's historical parking data, determine the candidate user's parking behavior characteristics, credit index, and collection sensitivity index; The pre-trained collection prediction model is used to process the parking feature data of each candidate user, and based on the output result of the collection prediction model, a target user recommended for collection among the candidate users is obtained, including: Input the parking behavior characteristics, credit index and collection sensitivity index of each candidate user into a pre-trained collection prediction model to obtain the collection success rate of each candidate user output by the collection prediction model; wherein the collection prediction model is trained based on the parking behavior characteristics, credit index and collection sensitivity index of each sample user, and sample labels representing the number of historical collection successes of each sample user; Based on the collection success rate of each candidate user, determine the target user recommended for collection from each candidate user; The method of determining, for each candidate user, the parking behavior characteristics, credit index, and collection sensitivity index of the candidate user based on the historical parking data of the candidate user includes: For each candidate user, the historical parking data of the candidate user is collected to obtain the parking behavior characteristics of the candidate user; Calculate the payment deferral index of the candidate user based on the departure time, payment time and actual payment time of each parking behavior of the candidate user, and calculate the credit index of the candidate user based on the payment deferral index of the candidate user; When no collection is made for the candidate user, determining the collection sensitivity index of the candidate user to be a first value; When the number of successful payment collection in the history of the candidate user is 0, determining the payment collection sensitivity index of the candidate user to be a second value; When the number of successful historical collection requests for the candidate user is greater than 0, the collection sensitivity index of the candidate user is calculated based on the actual payment time, collection time and preset payment time of each parking behavior of the candidate user that is being collected; wherein the preset payment time refers to the time between the vehicle departure time and the payment time.

2. The method according to claim 1, characterized in that The pre-trained collection prediction model is used to process the parking feature data of each candidate user, and based on the output result of the collection prediction model, a target user recommended for collection among the candidate users is obtained, including: The parking feature data of each candidate user is input into a pre-trained collection prediction model to obtain a target user for collection recommended by the collection prediction model; wherein the collection prediction model is trained based on the parking feature data of each sample user and a sample label indicating whether the collection for each sample user is successful.

3. The method according to claim 1, characterized in that: After determining the parking behavior characteristics, credit index and collection sensitivity index of each candidate user based on the historical parking data of the candidate user, the method further includes: Based on the parking behavior characteristics of each candidate user, each candidate user is grouped to obtain multiple candidate user groups; The step of inputting the parking behavior characteristics, credit index and collection sensitivity index of each candidate user into a pre-trained collection prediction model to obtain the collection success rate of each candidate user output by the collection prediction model includes: For each candidate user group, the parking behavior characteristics, credit index and collection sensitivity index of the candidate users in the group are input into the collection prediction model corresponding to the candidate user group, and the collection success rate of the candidate users in the group output by the collection prediction model is obtained; wherein the collection prediction model corresponding to the candidate user group is obtained by training based on the parking behavior characteristics, credit index and collection sensitivity index of each sample user of the corresponding sample user group, and sample labels representing the number of historical collection successes of each sample user of the sample user group; The step of determining a target user for recommended payment collection from among the candidate users based on the payment collection success rate of each candidate user includes: Based on the collection success rate of each group of candidate users, target users recommended for collection are determined from each group of candidate users.

4. The method according to claim 3, characterized in that: The step of determining the target user recommended for payment collection from each group of candidate users based on the payment collection success rate of each group of candidate users includes: For each candidate user group, based on the number of candidate users in the group and the prediction accuracy of the collection prediction model corresponding to the candidate user group for different numbers of users, the number of recommended collection users in the group of candidate users is determined as the first number; wherein the prediction accuracy is determined by predicting the collection success rate of the corresponding test user group based on the collection prediction model corresponding to the candidate user group; According to the target sorting result of the collection success rate of the group of candidate users from high to low, a first number of candidate users are determined from the group of candidate users to obtain the recommended collection users in the candidate user group; The recommended payment reminder users in each candidate user group are obtained to obtain the recommended payment reminder target users.

5. The method according to claim 4, characterized in that Methods for determining the prediction accuracy of the collection prediction model corresponding to the candidate user group for different numbers of users include: For each test user group, the parking behavior characteristics, credit index and collection sensitivity index of the test user group are input into the corresponding collection prediction model to obtain the collection success rate of the test user group; The collection success rates of the group of test users are sorted from high to low to obtain a first test sorting result of the group of test users; Sort the test users in this group by the number of successful payment collections in history from high to low, and use this as the second test sorting result for this group of test users; The first second number of test users in the second test sorting result are compared with the first second number of test users in the first test sorting result to determine the prediction accuracy of the collection prediction model corresponding to the candidate user group corresponding to the test user group for the first second number of users.

6. The method according to claim 3, characterized in that The parking behavior characteristics include: the candidate user's historical arrears amount and the number of successful collection requests in history; The method of grouping the candidate users based on the parking behavior characteristics of the candidate users to obtain multiple candidate user groups includes: Candidate users whose historical arrears amounts belong to the same amount range and whose historical successful collection times belong to the same number range are divided into the same candidate user group; wherein different candidate user groups correspond to different amount ranges and number ranges; the historical arrears amounts of the candidate user group and the corresponding sample user group belong to the same amount range, and the historical successful collection times belong to the same number range.

7. The method according to claim 1, characterized in that After determining the target user recommended for payment collection from the candidate users based on the payment collection success rate of each candidate user, the method further includes: Target users whose collection success rates are in the same success rate range are divided into a target user group, thereby obtaining multiple target user groups; For each target user group, based on the success rate interval to which the collection success rate of the target users in the group belongs and the corresponding relationship between the preset success rate interval and the collection frequency, determine the collection frequency of the target users in the group; The target users in this group are urged to pay according to their frequency of reminders.

8. The method according to any one of claims 1 to 7, characterized in that: The historical parking data includes at least one of the following: vehicle entry data, vehicle exit data, payment data and payment reminder data of the candidate user; The vehicle entry data includes: the parking lot identification and type; the vehicle exit data includes: parking time and order fee; the payment data includes: payment method and payment type; the collection data includes: collection amount and collection amount.

9. The method according to any one of claims 1 to 7, characterized in that: The collection prediction model is an extreme gradient boosting XGBoost model; or a gradient boosting decision tree GBDT model.

10. A model training method, characterized in that: The method comprises: Obtain historical parking data of each sample user in a historical time period; Generate training samples based on the historical parking data of each sample user; wherein the training samples include: parking behavior characteristics, credit index and collection sensitivity index of each sample user, and sample labels indicating the number of successful collections in history of each sample user; The parking behavior characteristics, credit index and collection sensitivity index of each sample user are input into the collection prediction model of the initial structure to obtain the collection success rate of each sample user; Based on the collection success rate of each sample user and the sample label indicating the number of successful collections in history of each sample user, the model parameters of the collection prediction model of the initial structure are adjusted to obtain a trained collection prediction model; For each sample user, the parking behavior characteristics of the sample user are: obtained by statistically analyzing the historical parking data of the sample user; The credit index of the sample user is calculated based on the payment deferral index of the sample user; the payment deferral index of the sample user is calculated based on the departure time, payment time and actual payment time of each parking behavior of the sample user; If the sample user has not been urged to pay, the collection sensitivity index of the sample user is the first value; If the number of successful collection requests in the history of the sample user is 0, the collection sensitivity index of the sample user is the second value; If the number of successful historical collection requests for the sample user is greater than 0, the collection sensitivity index of the sample user is calculated based on the actual payment time, collection time and preset payment time for each parking behavior of the sample user that is being collected; wherein the preset payment time refers to the time between the vehicle departure time and the payment time.

11. The method according to claim 10, characterized in that The generating of training samples based on the historical parking data of each sample user includes: For each sample user, the historical parking data of the sample user is collected to obtain the parking behavior characteristics of the sample user, and based on the historical parking data of the sample user, the credit index and collection sensitivity index of the sample user are calculated; For every two sample users, if the number of historical successful collections of the first sample user of the two sample users is greater than the number of historical successful collections of the second sample user, a positive sample is generated; wherein the positive sample includes the parking behavior characteristics, credit index and collection sensitivity index of the two sample users, and a first sample label indicating the number of historical successful collections of the two sample users; If the historical number of successful collections of the first sample user among the two sample users is less than the historical number of successful collections of the second sample user, a negative sample is generated; wherein the negative sample includes the parking behavior characteristics, credit index and collection sensitivity index of the two sample users, and a second sample label representing the historical number of successful collections of the two sample users.

12. The method according to claim 11, characterized in that After collecting historical parking data of each sample user to obtain parking behavior characteristics of the sample user, and calculating the credit index and collection sensitivity index of the sample user based on the historical parking data of the sample user, the method further includes: According to the historical arrears amount and the historical successful collection times in the parking behavior characteristics of the sample users, the sample users are grouped to obtain the sample user groups to which the sample users belong; wherein the historical arrears amount of the sample users belonging to the same sample user group belongs to the same amount range, and the historical successful collection times belong to the same number range; The parking behavior characteristics, credit index and collection sensitivity index of each sample user are input into the collection prediction model of the initial structure to obtain the collection success rate of each sample user, including: For each sample user group, the parking behavior characteristics, credit index and collection sensitivity index of the sample user group are input into the collection prediction model of the initial structure corresponding to the sample user group to obtain the collection success rate of the sample user group; The method adjusts the model parameters of the collection prediction model of the initial structure based on the collection success rate of each sample user and the sample label indicating the number of historical collection successes of each sample user to obtain a trained collection prediction model, including: Based on the collection success rate of the group of sample users and the sample labels representing the historical number of successful collections of the group of sample users, the model parameters of the collection prediction model of the initial structure corresponding to the sample user group are adjusted to obtain a trained collection prediction model corresponding to the sample user group.

13. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, for implementing the method described in any one of claims 1 to 9, or the method described in any one of claims 10 to 12, when executing a program stored in a memory.

14. A parking fee management system, characterized in that: The parking fee management system includes: a data collection module, a processor and a collection module; The data collection module is used to collect historical parking data of each candidate user in a historical time period; wherein the candidate user is a user who has outstanding parking fees in the parking lot; The processor is used to execute the method according to any one of claims 1 to 9 to obtain a target user for recommended payment collection; The collection module is used to collect payment from the target user.

Images