User payment behavior prediction method based on trend, reason and confidence coefficient

Abstract

The embodiment of the invention provides a user payment behavior prediction method based on trends, reasons and confidence. The technical problem that an existing game user payment behavior prediction method is large in prediction error can be solved. The user payment behavior prediction method based on the trend, the reason and the confidence coefficient comprises the following steps: dividing user data into payment behavior data and non-payment behavior data, and respectively training a first prediction model and a second prediction model; the first prediction model predicts a future payment behavior of the user by learning a past payment behavior trend. The second prediction model learns a mapping relation between a non-payment behavior and a user payment behavior from the perspective of reasons. The two prediction models predict the payment behavior from different perspectives, and the result is more accurate. In addition, the payment behavior of the user is predicted in combination with the model confidence degree, and the influence of the reliability degree of the prediction model on the payment behavior is considered.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method for predicting user payment behavior based on trends, causes and confidence levels. Background Art

[0002] Game user payment behavior prediction, by understanding the user's payment behavior, predicts whether the user will pay or purchase a product or service. Game user payment behavior prediction helps game platforms make smarter business decisions and strategic planning, such as improving game pricing strategies, game product features, and improving user experience and user retention.

[0003] Currently, the behavioral data and other relevant characteristics of game users are analyzed by manual screening and extraction of payment-related behavioral characteristics.

[0004] However, manual analysis is greatly influenced by the analysts’ subjective influence and cannot adaptively control the impact of different features on payment behavior, which can easily lead to unsatisfactory prediction results. Summary of the invention

[0005] The embodiment of the present application provides a method for predicting user payment behavior based on trends, causes and confidence levels, which can solve the technical problem of large prediction errors in existing methods for predicting game user payment behavior.

[0006] To achieve the above objectives, the embodiments of the present application adopt the following technical solutions: In a first aspect, an embodiment of the present application provides a method for predicting user payment behavior based on trends, causes and confidence, and the method for predicting user payment behavior based on trends, causes and confidence includes: dividing user data into first data and second data; the first data is the payment behavior data of each user, and the first data includes the payment data of each user every day, and the payment data includes time characteristics; the second data is the non-payment behavior data of each user, and the second data includes the non-payment data of each user every day, and the non-payment data includes multiple non-payment features; based on the time characteristics, encoding the first data to obtain first encoded data; periodically traversing the first encoded data to construct a first data set; based on the first loss function and the first data set, training a first prediction model; based on the non-payment features, encoding the second data to obtain second encoded data; periodically traversing the second encoded data to construct a second data set; training a second prediction model based on the second loss function and the second data set; based on the historical payment rate, confirming the first paying user set, the second paying user set and the third paying user set. user set; based on the first data set, establish a first evaluation data set, a second evaluation data set and a third evaluation data set corresponding to the first paying user set, the second paying user set and the third paying user set respectively; based on the second data set, establish a fourth evaluation data set, a fifth evaluation data set and a sixth evaluation data set corresponding to the first paying user set, the second paying user set and the third paying user set respectively; according to the first prediction model, obtain a first confidence based on the first evaluation data set, obtain a second confidence based on the second evaluation data set, and obtain a third confidence based on the third evaluation data set; according to the second prediction model, obtain a fourth confidence based on the fourth evaluation data set, obtain a fifth confidence based on the fifth evaluation data set, and obtain a sixth confidence based on the sixth evaluation data set; obtain a payment prediction result according to the first combined probability, the second combined probability or the third combined probability; the first combined probability includes the first confidence and the fourth confidence; the second combined probability includes the second confidence and the fifth confidence; the third combined probability includes the third confidence and the sixth confidence.

[0007] Based on the above description of the user payment behavior prediction method based on trends, causes and confidence provided in the embodiment of the present application, it can be known that the user payment behavior prediction method based on trends, causes and confidence includes dividing the user data into payment behavior data and non-payment behavior data, and training a first prediction model and a second prediction model respectively. The first prediction model predicts the user's future payment behavior by learning past payment behavior trends. The second prediction model learns the mapping relationship between non-payment behavior and user payment behavior from the perspective of causes. The two prediction models predict payment behavior from different perspectives, and the results are more accurate.

[0008] In addition, the present application designs and combines model confidence to predict user payment behavior, which takes into account the impact of the reliability of the prediction model itself on payment behavior.

[0009] In a feasible implementation of the first aspect, the first coded data includes a first coded element, a second coded element, a third coded element, a fourth coded element, a fifth coded element, a sixth coded element, a seventh coded element, an eighth coded element and a ninth coded element. With a seven-day period as a cycle, when executing the step of periodically traversing the first coded data to construct a first data set, the user payment behavior prediction method based on trends, causes and confidence also includes: using the first coded element, the second coded element, the third coded element, the fourth coded element, the fifth coded element, the sixth coded element and the seventh coded element as the first input data of the first data set; using the non-coded data corresponding to the eighth coded element as the first data label of the first input data; using the second coded element, the third coded element, the fourth coded element, the fifth coded element, the sixth coded element, the seventh coded element and the eighth coded element as the second input data of the first data set; using the non-coded data corresponding to the ninth coded element as the second data label of the first input data.

[0010] In a feasible implementation of the first aspect, the first loss function is a cross entropy loss function, and the first prediction model is established based on a Transformer model.

[0011] In a feasible implementation of the first aspect, the second coded data includes first non-payment data; the first non-payment data includes a first non-payment feature, a second non-payment feature, a third non-payment feature, a fourth non-payment feature, a fifth non-payment feature and a sixth non-payment feature; when executing the step of encoding the second data based on the non-payment feature to obtain the second coded data, the user payment behavior prediction method based on trends, causes and confidence also includes: based on the first non-payment feature, through a first preset condition, obtaining a first characteristic value; the first characteristic value includes a first upper limit value; based on the second non-payment feature, through a second preset condition, obtaining a second characteristic value; the second characteristic value includes a second upper limit value; based Based on the third non-payment feature, a third characteristic value is obtained through the third preset condition; the third characteristic value includes the third upper limit value; based on the fourth non-payment feature, a fourth characteristic value is obtained through the fourth preset condition; the fourth characteristic value includes the fourth upper limit value; based on the fifth non-payment feature, a fifth characteristic value is obtained through the fifth preset condition; the fifth characteristic value includes the fifth upper limit value; based on the sixth non-payment feature, a sixth characteristic value is obtained through the sixth preset condition; the sixth characteristic value includes the sixth upper limit value; wherein, the first upper limit value is less than the second upper limit value, the second upper limit value is less than the third upper limit value, the third upper limit value is less than the fourth upper limit value, the fourth upper limit value is less than the fifth upper limit value, and the fifth upper limit value is less than the sixth upper limit value.

[0012] In a feasible implementation of the first aspect, when executing the step of obtaining a payment prediction result based on the first combination probability, the second combination probability or the third combination probability, the user payment behavior prediction method based on trends, causes and confidence also includes: confirming the first historical payment behavior and historical non-payment behavior of the user to be predicted; inputting the first historical payment behavior into the first prediction model to obtain a first predicted payment rate; inputting the historical non-payment behavior into the second prediction model to obtain a second predicted payment rate; confirming the second historical payment behavior of the user to be predicted; obtaining the payment rate to be tested based on the second historical payment behavior; confirming the target payment level based on the payment rate to be tested; the target payment level is the first paying user set, the second paying user set or the third paying user set; if the target payment level is the first paying user set, obtaining the payment prediction result based on the first combination probability.

[0013] In a feasible implementation of the first aspect, before executing the step of dividing the user data into first data and second data, the user payment behavior prediction method based on trends, reasons and confidence also includes: obtaining the user's historical behavior log data in the game platform, and cleaning and preprocessing the historical behavior log data to obtain user data.

[0014] In a feasible implementation of the first aspect, the second loss function is a cross entropy loss function, and the second prediction model is established based on a Transformer model.

[0015] In a feasible implementation of the first aspect, the encoding formula for calculating the first encoded data includes: ; in, Expressed as The day of the week corresponding to the day, the value is 1 to 7; Expressed as Users in the past Is there any payment behavior on that day?

[0016] In a second aspect, an embodiment of the present application provides a user payment behavior prediction system based on trends, reasons and confidence, and the user payment behavior prediction system based on trends, reasons and confidence includes: at least one processor; a memory communicatively connected to at least one processor; wherein the memory stores instructions executable by at least one processor, and the instructions are executed by at least one processor so that at least one processor can execute the method provided in the first aspect.

[0017] The user payment behavior prediction system based on trends, reasons and confidence divides user data into payment behavior data and non-payment behavior data by executing the method provided in the first aspect, and trains the first prediction model and the second prediction model respectively. The first prediction model predicts the user's future payment behavior by learning the past payment behavior trend. The second prediction model learns the mapping relationship between non-payment behavior and user payment behavior from the perspective of cause. The two prediction models predict payment behavior from different perspectives, and the results are more accurate.

[0018] In a third aspect, an embodiment of the present application provides a computer-readable medium having computer program instructions stored thereon, and the computer program instructions can be executed by a processor to implement the method provided in the first aspect.

[0019] The computer program instructions in the computer-readable medium divide the user data into payment behavior data and non-payment behavior data by implementing the method provided in the first aspect, and train the first prediction model and the second prediction model respectively. The first prediction model predicts the user's future payment behavior by learning the past payment behavior trend. The second prediction model learns the mapping relationship between non-payment behavior and user payment behavior from the perspective of cause. The two prediction models predict payment behavior from different perspectives, and the results are more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of the structure of a user payment behavior prediction system based on trends, causes and confidence levels provided in an embodiment of the present application; Figure 2 A flowchart of a method for predicting user payment behavior based on trends, causes and confidence levels provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The technical solution in the embodiment of the present invention will be described below in conjunction with the drawings in the embodiment of the present invention. In the description of the embodiment of the present invention, unless otherwise specified, "multiple" means two or more than two. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0022] In addition, in order to facilitate the clear description of the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, the words "first", "second" and the like are used to distinguish the same items or similar items with substantially the same functions and effects. Those skilled in the art will understand that the words "first", "second" and the like do not limit the quantity and execution order, and the words "first", "second" and the like do not necessarily limit the difference. At the same time, in the embodiments of the present invention, the words "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or design. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way for easy understanding.

[0023] The principles and features of the present application are described below. The examples given are only used to explain the present application and are not used to limit the scope of the present application.

[0024] The embodiment of the present application provides a user payment behavior prediction system based on trends, causes and confidence levels, which can execute the user payment behavior prediction method based on trends, causes and confidence levels provided in the embodiment of the present application. Figure 1 A schematic diagram of the structure of a user payment behavior prediction system based on trends, causes and confidence levels provided in an embodiment of the present application.

[0025] like Figure 1 As shown, the user payment behavior prediction system 001 based on trends, reasons and confidence includes at least one processor 011 and a memory 012 communicatively connected to the at least one processor; wherein the memory 012 stores instructions executable by the at least one processor 011, and the instructions are executed by the at least one processor 011 so that the at least one processor 011 can execute the user payment behavior prediction method based on trends, reasons and confidence provided in an embodiment of the present application.

[0026] Figure 2 The following is a flow chart of a method for predicting user payment behavior based on trends, causes and confidence levels provided in an embodiment of the present application. Figure 2 As shown, in some embodiments, the user payment behavior prediction method based on trend, reason and confidence includes the following steps: S1, dividing user data into first data and second data.

[0027] The first data is the payment behavior data of each user. The first data includes the payment data of each user every day. The payment data includes time characteristics. For example, the payment behavior data in the user data is used as the first data. , , where the subscript Expressed as the total number of users, Expressed as The payment behavior data of users in the past month. , where the subscript Expressed as the total number of days in the past month, Expressed as Users in the past Whether there is payment behavior on the day. Users in the past If there is payment behavior on the day The value of is 1. Users in the past If there is no payment behavior on the day, The value of is 0. It can be seen that payment behavior is associated with user information ( users) and time information ( days), i.e. the paid data includes time characteristics.

[0028] The second data is the non-payment behavior data of each user. The second data includes the non-payment data of each user every day. The non-payment data includes multiple non-payment features. In one implementation, the non-payment behavior data in the user data is used as the second data. , , where the subscript Expressed as the total number of users, Expressed as Non-paying behavior data of users in the past month. , where the subscript Expressed as the total number of days in the past month, Expressed as Users in the past days of free data. A number of non-payment features may be included. In one example, Six non-paying features can be included, ,in, Expressed as Users in the past Number of mall visits per day, Expressed as Users in the past Number of times products are visited in the mall per day, Expressed as Users in the past The number of operations on the recharge interface per day, Expressed as Users in the past The time spent in the mall per day, Expressed as Users in the past Total number of games per day, Expressed as Users in the past Number of logins per day.

[0029] In the embodiment of the present application, the user's historical behavior is divided into payment behavior data and non-payment behavior data, where the historical payment behavior represents the user's past payment trend. Constructing a first prediction model based on trends and introducing time coding features will help to mine the periodic correlation of user payment behavior over time. Non-payment behavior data consists of some behaviors related to user payment, which are often the inducement for user payment behavior, such as the user's effective stay time in the mall, the number of clicks on the product, the frequency of visits, etc., which can all be used to predict the probability of user payment. Therefore, constructing a second prediction model based on causes will help to mine the correlation between the user's non-payment behavior and payment behavior, and predict user payment from multiple different dimensions with greater accuracy.

[0030] The user data is divided into paid behavior data and non-paid behavior data to prepare for training the first prediction model and the second prediction model respectively.

[0031] In some embodiments, before executing step S1, the user payment behavior prediction method based on trend, reason and confidence further includes the following steps: S100, obtaining historical behavior log data of the user in the game platform, and cleaning and preprocessing the historical behavior log data to obtain user data.

[0032] In some embodiments, the cleaning method includes deleting the user data of users whose registration time is less than one month. The cleaning method may also include deleting the user data of users who have not logged in since registration, or have logged in less than 3 times per month. The cleaning method may also include intercepting the user's data of the most recent month for users whose registration time exceeds one year. For users whose registration time is less than one month, all the user's data is retained.

[0033] S2, encoding the first data based on the time characteristic to obtain first encoded data.

[0034] In some embodiments, the encoding formula for calculating the first encoded data includes: ; in, Expressed as The day of the week corresponding to the day, with a value ranging from 1 to 7. Expressed as Users in the past Is there any payment behavior on that day?

[0035] For example, A user paid on the first day of the past month, and the first day corresponds to a Wednesday, then The value of is 3, The value of is 1, The corresponding first coded data .

[0036] Using the above method to traverse, in one example, by The payment behavior coding data of the user's introduction time information in the past month is composed of ,but .in, The value of is 30.

[0037] Continue to traverse in the above manner to obtain the payment behavior data of all users after encoding. Then the first encoded data .

[0038] S3, periodically traverse the first encoded data to construct a first data set.

[0039] In some embodiments, when multiple users are involved, when executing step S3, arrive Traversal Middle Payment behavior coding data of each user , get the first data set In one implementation, traverse the first The payment behavior coding data of each user for seven consecutive days is the input data of the first data set. As the label of the input data, the label is non-encoded payment behavior data, and its value is 0 or 1.

[0040] Exemplarily, the first coded data Includes the first encoding element , the second coding element , the third coding element , the fourth coding element , the fifth coding element , the sixth coding element , the seventh coding element , the eighth coding element and the ninth coding element , taking seven days as a cycle, when executing the step of periodically traversing the first coded data to construct the first data set, the user payment behavior prediction method based on trend, reason and confidence also includes the following steps: S31, taking a first coding element, a second coding element, a third coding element, a fourth coding element, a fifth coding element, a sixth coding element and a seventh coding element as first input data of a first data set.

[0041] First encoding element , the second coding element , the third coding element , the fourth coding element , the fifth coding element , the sixth coding element , the seventh coding element , as the first input data.

[0042] S32, using the non-coded data corresponding to the eighth coding element as the first data label of the first input data.

[0043] The value of the non-coded data corresponding to the eighth coded element is 0 or 1.

[0044] S33, taking the second coding element, the third coding element, the fourth coding element, the fifth coding element, the sixth coding element, the seventh coding element and the eighth coding element as the second input data of the first data set.

[0045] Second encoding element , the third coding element , the fourth coding element , the fifth coding element , the sixth coding element , the seventh coding element , the eighth coding element , as the second input data.

[0046] S34, using the non-coded data corresponding to the ninth coded element as the second data label of the first input data.

[0047] The value of the non-coded data corresponding to the ninth coded element is 0 or 1.

[0048] It is understandable that steps S31 to S34 are merely exemplary of two cyclic operations of periodic traversal, and do not constitute a limitation on the number of cyclic operations of periodic traversal. The application does not limit the number of cyclic operations of periodic traversal. Also, the cyclic operations of periodic traversal can refer to steps S31 to S34.

[0049] S4: training a first prediction model based on the first loss function and the first data set.

[0050] In some embodiments, the first loss function is a cross entropy loss function, and the first prediction model Based on the Transformer model.

[0051] In one implementation, the first loss function The calculation formula includes: ; in, Represented as from the first data set The number of data sampled in ; Represented as sampling The first If the data The data is The payment behavior data of each user; ; Expressed as Enter the first prediction model The output result obtained; yes The corresponding true label.

[0052] Based on the above dataset And the loss function Train the first prediction model until the loss function value no longer decreases, and obtain the optimized first prediction model .

[0053] In this way, the first prediction model predicts the user's future payment behavior by learning past payment behavior trends.

[0054] S5, based on the non-payment feature, encode the second data to obtain second encoded data.

[0055] Non-payment features include behavioral data that can influence user payment. For example, non-payment features may include the number of times a user visits the mall, the number of times a user visits products in the mall, the number of operations on the recharge interface, and the effective stay time in the mall (effective stay time refers to the time during which other operations are performed during the stay). Six non-paying features can be included, ,in, It is represented as the first non-paid feature, Users in the past Number of mall visits per day; Indicated as the second non-paid feature, Users in the past Number of times products are visited in the mall per day; Indicated as the third non-paid feature, Users in the past Number of operations on the recharge interface per day; It is indicated as the fourth non-paid feature, Users in the past The duration of stay in the mall per day; It is indicated as the fifth non-paid feature, Users in the past Total number of games played in a day; It is indicated as the sixth non-paid feature, Users in the past Number of logins per day.

[0056] When executing step S5, the user payment behavior prediction method based on trend, reason and confidence also includes the following steps: S51, based on the first non-payment feature, obtain a first feature value through a first preset condition.

[0057] The first characteristic value includes a first upper limit value.

[0058] In some embodiments, the first preset condition includes calculating the formula , and get the first eigenvalue The calculated value of, and, according to the first upper limit value and the first characteristic value The calculated value of confirms the first eigenvalue The final result.

[0059] In one implementation, the first eigenvalue The value range of is 0-100, and the first upper limit is 100. That is, in the first eigenvalue When the calculated value is greater than 100, the first eigenvalue The final result is 100. For example, the first non-paid feature is 110, the first eigenvalue The calculated value of is 110, then the first eigenvalue The final result is 100.

[0060] S52: Based on the second non-payment feature, a second feature value is obtained through a second preset condition.

[0061] The second characteristic value includes a second upper limit value.

[0062] In some embodiments, the second preset condition includes calculating the formula , and obtain the second eigenvalue The calculated value of, and, according to the second upper limit value and the second characteristic value The calculated value of confirms the second eigenvalue The final result.

[0063] In one implementation, the second eigenvalue The value range of is 100-200, and the second upper limit is 200. That is to say, in the second eigenvalue When the calculated value is greater than 200, the second eigenvalue The final result is 200. For example, the second non-paid feature is 110, the second eigenvalue The calculated value of is 210, then the second eigenvalue The final result is 200.

[0064] S53: Based on the third non-payment feature and through a third preset condition, a third feature value is obtained.

[0065] The third characteristic value includes a third upper limit value.

[0066] In some embodiments, the third preset condition includes calculating the formula , and get the third eigenvalue The calculated value of, and, according to the third upper limit value and the third characteristic value The calculated value of confirms the third eigenvalue The final result.

[0067] In one implementation, the third eigenvalue The value range of is 200-400, and the third upper limit is 400. That is, at the third eigenvalue When the calculated value is greater than 400, the third eigenvalue The final result is 400. For example, the third non-paid feature is 210, the third eigenvalue The calculated value of is 410, so the third eigenvalue The final result is 400.

[0068] S54, based on the fourth non-payment feature and through a fourth preset condition, obtain a fourth feature value.

[0069] The fourth characteristic value includes a fourth upper limit value.

[0070] In some embodiments, the fourth preset condition includes calculating the formula , and obtain the fourth eigenvalue The calculated value of, and, according to the fourth upper limit value and the fourth characteristic value The calculated value of confirms the fourth eigenvalue The final result.

[0071] In one implementation, the fourth eigenvalue The value range of is 400-500, and the fourth upper limit is 500. That is to say, in the fourth eigenvalue When the calculated value is greater than 500, the fourth eigenvalue The final result is 500. For example, the fourth non-paid feature is 210, the fourth eigenvalue The calculated value of is 610, so the fourth eigenvalue The final result is 500.

[0072] It is understandable that the effective stay time in the mall is measured in minutes, and any fraction of a minute will be calculated as one minute.

[0073] In some embodiments, the fourth non-payment feature The time limit is 210. If it exceeds 100 minutes, it will be calculated as 100 minutes.

[0074] S55, based on the fifth non-payment feature and through the fifth preset condition, obtain a fifth feature value.

[0075] The fifth characteristic value includes a fifth upper limit value.

[0076] In some embodiments, the fifth preset condition includes calculating the formula , and get the fifth eigenvalue The calculated value of, and, according to the fifth upper limit value and the fifth characteristic value The calculated value of confirms the fifth eigenvalue The final result.

[0077] In one implementation, the fifth eigenvalue The value range of is 500-1500, and the fifth upper limit is 1500. That is to say, in the fifth eigenvalue When the calculated value is greater than 1500, the fifth eigenvalue The final result is 1500. For example, the fifth non-paid feature is 1110, the fifth eigenvalue The calculated value of is 1610, so the fifth eigenvalue The final result is 1500.

[0078] S56, based on the sixth non-payment feature and through the sixth preset condition, obtain a sixth feature value.

[0079] The sixth characteristic value includes a sixth upper limit value.

[0080] In some embodiments, the sixth preset condition includes calculating the formula , and get the sixth eigenvalue The calculated value of, and, according to the sixth upper limit value and the sixth characteristic value The calculated value of confirms the sixth eigenvalue The final result.

[0081] In one implementation, the sixth eigenvalue The value range of is 1500-1600, and the sixth upper limit is 1600. That is to say, at the sixth eigenvalue When the calculated value is greater than 1600, the sixth eigenvalue The final result is 1600. For example, the sixth non-paid feature is 1110, the sixth eigenvalue The calculated value of is 2610, so the sixth eigenvalue is The final result is 1600.

[0082] It should be noted that the first upper limit value is smaller than the second upper limit value, the second upper limit value is smaller than the third upper limit value, the third upper limit value is smaller than the fourth upper limit value, the fourth upper limit value is smaller than the fifth upper limit value, and the fifth upper limit value is smaller than the sixth upper limit value.

[0083] By executing steps S51 to S56, the operation of encoding the second data based on the non-payment feature to obtain the second encoded data is completed. The obtained second encoded data .

[0084] The purpose of encoding is to convert user data into dictionary data that can be recognized by the Transformer model. Since many values ​​in the original data are repeated, the meaning of these values ​​cannot be distinguished by directly inputting them into the model. For example, the number of product visits is 2, and the number of recharge operations is also 2. These two values ​​are the same (there may even be other values ​​of 2), and the model cannot distinguish which data has the value of 2. However, after encoding, the value of the number of product visits is 100+2=102, and the value of the number of recharge operations is 200+2=202. Although the actual values ​​are the same, the results after encoding are different, and the model will learn the meaning of these different values. Therefore, encoding requires that all values ​​are not repeated and each value has a different meaning.

[0085] S6, periodically traverse the second encoded data to construct a second data set.

[0086] In some embodiments, when multiple users are involved, when executing step S6, arrive Traversal Middle Non-payment behavior coding data of users , construct the second data set .

[0087] In one implementation, traverse the first Non-payment behavior coding data of users for seven consecutive days is the input data of the second data set. As a label of the input data, the label is non-encoded non-payment behavior data, and takes a value of 0 or 1.

[0088] It is understandable that the method of periodic traversal can be referred to step S3, which will not be described in detail here.

[0089] S7, training a second prediction model based on the second loss function and the second data set.

[0090] In some embodiments, the second loss function is a cross entropy loss function, and the second prediction model Based on the Transformer model.

[0091] In one implementation, the calculation formula of the second loss function includes: ; in, Represented as from the second data set The number of data sampled in ; Represented as sampling The first If the data The data is Non-payment behavior data of each user; ; Expressed as Enter the second prediction model The output result obtained; yes The corresponding true label.

[0092] Based on the second dataset and the second loss function The second prediction model is trained until the loss function value no longer decreases, thereby obtaining an optimized second prediction model.

[0093] In this way, the second prediction model learns the mapping relationship between non-paying behavior and user paying behavior from the perspective of cause.

[0094] S8, based on the historical payment rates, confirm the first paying user set, the second paying user set and the third paying user set.

[0095] In some embodiments, the historical payment rate of each user is calculated based on the historical payment behavior data in the historical behavior log data. It can be understood that the historical behavior log data includes data of multiple users.

[0096] In one implementation, the calculation formula of the historical payment rate includes: ; in, Expressed as Users; It represents the total number of days in the past month; m represents the mth day in the past month; Expressed as The historical payment rate of each user; Expressed as Users in the past Is there any payment behavior on that day?

[0097] Use the above method to traverse the first data , get the payment rate of all users : .

[0098] Based on the historical payment rate of any user, the user is divided into different paying user sets. Different paying user sets are used to characterize different payment frequencies of users. Exemplarily, the payment frequency may include high frequency payment, normal payment, and low frequency payment, a total of three categories. High frequency payment corresponds to the first paying user set. Normal payment corresponds to the second paying user set. Low frequency payment corresponds to the third paying user set.

[0099] In one implementation, all users are charged a fee Sort from high to low, all the sorted Each user uses , ,..., Indicates that the sorted The users are divided into three equal parts, corresponding to high-frequency (high) payment, normal payment and low-frequency (low) payment. . Normal paying users Low-frequency paying users .

[0100] S9, based on the first data set, establish a first evaluation data set, a second evaluation data set and a third evaluation data set corresponding to the first paying user set, the second paying user set and the third paying user set respectively.

[0101] First Dataset Including the first encoded data and its corresponding label. Find all high paying users Corresponding data, construct the first evaluation data set In the first data set Find all normal paying users Corresponding data, construct the second evaluation data set In the first data set Find all low-frequency paying users Corresponding data, construct the third evaluation data set .

[0102] S10, based on the second data set, establishing a fourth evaluation data set, a fifth evaluation data set and a sixth evaluation data set corresponding to the first paying user set, the second paying user set and the third paying user set respectively.

[0103] Second dataset Including the second encoded data and its corresponding label. Find all high paying users The corresponding data constructs the fourth evaluation data set In the second data set Find all normal paying users The corresponding data constructs the fifth evaluation data set In the second data set Find all low-frequency paying users The corresponding data constructs the sixth evaluation data set .

[0104] S11, according to the first prediction model, obtain a first confidence based on the first evaluation data set, obtain a second confidence based on the second evaluation data set, and obtain a third confidence based on the third evaluation data set.

[0105] According to the first prediction model ,enter , get the first confidence .

[0106] In one implementation, the first confidence The calculation formula includes: .

[0107] According to the first prediction model ,enter , get the second confidence .

[0108] In one implementation, the second confidence The calculation formula includes: .

[0109] According to the first prediction model ,enter , and get the third confidence level .

[0110] In one implementation, the third confidence level The calculation formula includes: .

[0111] in, Represented as a dataset The total number of data; Indicates that the first prediction model is for the input data Output prediction results; represents an indicator function, for example, Indicates that when satisfied hour, ,otherwise .

[0112] S12, according to the second prediction model, obtain a fourth confidence based on the fourth evaluation data set, obtain a fifth confidence based on the fifth evaluation data set, and obtain a sixth confidence based on the sixth evaluation data set.

[0113] According to the second prediction model ,enter , and get the fourth confidence level .

[0114] In one implementation, the fourth confidence The calculation formula includes: .

[0115] According to the second prediction model ,enter , get the fifth confidence level .

[0116] In one implementation, the fifth confidence level The calculation formula includes: .

[0117] According to the second prediction model ,enter , and get the sixth confidence level .

[0118] In one implementation, the sixth confidence level The calculation formula includes: .

[0119] S13, obtaining a payment prediction result according to the first combination probability, the second combination probability or the third combination probability.

[0120] The first combined probability includes a first confidence level and a fourth confidence level.

[0121] The second combined probability includes the second confidence level and the fifth confidence level.

[0122] The third combined probability includes the third confidence level and the sixth confidence level.

[0123] In some embodiments, when executing step S13, the user payment behavior prediction method based on trend, reason and confidence level further includes the following steps: S131, confirming the first historical payment behavior and the historical non-payment behavior of the user to be predicted.

[0124] In one implementation, the user to be predicted may be The first historical payment behavior can be the coded data of the user to be predicted in the past seven days, including The historical non-paying behavior can be the coded data of the user to be predicted in the past seven days, including .

[0125] S132: Input the first historical payment behavior into the first prediction model to obtain a first predicted payment rate.

[0126] In one implementation, the first historical payment behavior Input into the first prediction model to obtain the first predicted payment rate .

[0127] S133, inputting the historical non-paying behavior into the second prediction model to obtain a second predicted payment rate.

[0128] In one implementation, historical non-paying behaviors Input into the second prediction model to obtain the second predicted payment rate .

[0129] S134, confirming the second historical payment behavior of the user to be predicted.

[0130] In one implementation, the second historical payment behavior may be The user's payment history in the past month (can be 30 days), including .

[0131] S135, obtaining a payment rate to be tested based on the second historical payment behavior.

[0132] In one implementation, based on the second historical payment behavior , get the payment rate to be tested .

[0133] S136: confirm the target payment level based on the payment rate to be tested.

[0134] The target payment level is the first paying user set, the second paying user set, or the third paying user set.

[0135] In one implementation, based on The value of The user belongs to a high-frequency paying user (i.e., the first paying user set), a normal paying user (i.e., the second paying user set), or a low-frequency paying user (i.e., the third paying user set). It is understandable that when the user is divided into different paying frequency user sets, the division result of step S8 can be used as a basis, which will not be repeated here.

[0136] S137: If the target payment level is the first paying user set, a payment prediction result is obtained according to the first combination probability.

[0137] If the user to be predicted belongs to the first paying user set, then according to the first confidence , the fourth confidence level , the first predicted payment rate and the second predicted payment rate Get paid prediction results.

[0138] In one implementation, the payment prediction result may be a probability value, and the calculation formula of the payment prediction result includes: ; in, It is expressed as the first predicted payment rate; Expressed as the second predicted payment rate; It is expressed as the first confidence level; Indicated as the fourth confidence level.

[0139] For example, suppose the model confidence is =0, =0, then the final R=0.5, which means that the probability of the user paying is 0.5, that is, it is difficult to tell whether the user will pay in the future. This is reasonable, because a model confidence of 0 means that the model is unreliable, and we cannot rely on the model results to judge whether the user will pay in the future. On the contrary, if the model confidence is 1, that is, the model prediction results are very reliable, then the final result R will directly depend on the model prediction results.

[0140] In some other embodiments, if the user to be predicted belongs to the second paying user set, the user is predicted based on the second confidence level. , the fifth confidence level , the first predicted payment rate and the second predicted payment rate Get paid prediction results.

[0141] The calculation formula for the payment prediction result includes: ; In some other embodiments, if the user to be predicted belongs to the third paying user set, the user is predicted based on the third confidence level. , the sixth confidence level , the first predicted payment rate and the second predicted payment rate Get paid prediction results.

[0142] The calculation formula for the payment prediction result includes: ; The advantage of this formula for calculating the probability of predicted payment behavior is that it improves the accuracy of model prediction by combining the predicted payment rate and confidence.

[0143] This application mines the trend characteristics and cause characteristics that affect user payment behavior, and constructs the first prediction model and the second prediction model of user payment behavior based on these characteristics. The first prediction model considers the impact of users' past payment behavior on future payment. Its core is to predict future payment behavior trends through the trends of users' past payment behavior. The second prediction model considers the impact of users' non-payment behavior on future payment behavior from the perspective of causes. The two prediction models predict payment behavior from different perspectives, and the results are more accurate, reducing the model's misjudgment rate. In addition, this application designs and combines model confidence to predict user payment behavior, which takes into account the impact of the prediction model's own reliability on payment behavior.

[0144] Furthermore, through the calculation method of model prediction confidence, the prediction confidence considers the reliability of the first prediction model and the second prediction model in predicting the payment behavior of different users from the perspective of model prediction ability. The model prediction confidence is divided into groups according to the similarity of user payment behavior, and the model prediction confidence of each group of users paying and not paying is calculated based on the first prediction model and the second prediction model. Based on the evaluation data set constructed based on three categories, the first prediction model of user payment behavior based on trends and the second prediction model of user payment behavior based on reasons are used for prediction. Based on the prediction results and the historical actual payment behavior results, the model prediction confidence of the first prediction model and the second prediction model in different categories are calculated respectively. Finally, the results of the first prediction model, the second prediction model and the model prediction confidence are combined to predict the user's future payment behavior, which can effectively improve the accuracy of payment behavior prediction.

[0145] As mentioned above, this application combines machine learning and fully considers the user payment behavior trend, payment reasons, and the impact of the model prediction confidence level on the user payment behavior, which can effectively improve the accuracy of payment behavior prediction.

[0146] Based on the same application concept, a user payment behavior prediction system based on trends, causes and confidence is also provided in an embodiment of the present application. The method corresponding to the user payment behavior prediction system based on trends, causes and confidence may be the user payment behavior prediction method based on trends, causes and confidence in the aforementioned embodiment, and its principle of solving the problem is similar to that of the method. The user payment behavior prediction system based on trends, causes and confidence provided in an embodiment of the present application includes: at least one processor. And a memory communicatively connected to the at least one processor. Wherein, the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the methods and / or technical solutions of the aforementioned multiple embodiments of the present application.

[0147] Another embodiment of the present application further provides a computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of the present application described above.

[0148] Specifically, the present embodiment may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by an instruction execution system, device, or device or used in combination therewith.

[0149] Computer readable signal media may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer readable program code. Such propagated data signals may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than a computer readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0150] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0151] Computer program code for performing the operation of the present application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect through the Internet).

[0152] The flow chart or block diagram in the accompanying drawings shows the possible architecture, function and operation of the equipment, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated system for hardware that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0153] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0154] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or page components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0155] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0156] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0157] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

[0159] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in a device claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

Claims

1. A method for predicting user payment behavior based on trends, causes and confidence, characterized in that: include: Divide the user data into first data and second data; the first data is the payment behavior data of each user, the first data includes the payment data of each user every day, and the payment data includes time characteristics; the second data is the non-payment behavior data of each user, the second data includes the non-payment data of each user every day, and the non-payment data includes multiple non-payment characteristics; Based on the time characteristic, encoding the first data to obtain first encoded data; Periodically traverse the first encoded data to construct a first data set; Training a first prediction model based on a first loss function and the first data set; Based on the non-payment feature, encoding the second data to obtain second encoded data; Periodically traverse the second encoded data to construct a second data set; Based on the second loss function and the second data set, training a second prediction model; Based on the historical payment rates, confirming the first paying user set, the second paying user set, and the third paying user set; Based on the first data set, establishing a first evaluation data set, a second evaluation data set, and a third evaluation data set corresponding to the first paying user set, the second paying user set, and the third paying user set, respectively; Based on the second data set, establishing a fourth evaluation data set, a fifth evaluation data set and a sixth evaluation data set corresponding to the first paying user set, the second paying user set and the third paying user set respectively; According to the first prediction model, obtaining a first confidence level based on the first evaluation data set, obtaining a second confidence level based on the second evaluation data set, and obtaining a third confidence level based on the third evaluation data set; According to the second prediction model, a fourth confidence level is obtained based on the fourth evaluation data set, a fifth confidence level is obtained based on the fifth evaluation data set, and a sixth confidence level is obtained based on the sixth evaluation data set; A payment prediction result is obtained according to a first combined probability, a second combined probability or a third combined probability; the first combined probability includes the first confidence level and the fourth confidence level; the second combined probability includes the second confidence level and the fifth confidence level; the third combined probability includes the third confidence level and the sixth confidence level.

2. The user payment behavior prediction method based on trend, cause and confidence according to claim 1 is characterized in that: The first coded data includes a first coded element, a second coded element, a third coded element, a fourth coded element, a fifth coded element, a sixth coded element, a seventh coded element, an eighth coded element and a ninth coded element. With seven days as a cycle, when executing the step of periodically traversing the first coded data to construct a first data set, the user payment behavior prediction method based on trend, reason and confidence also includes: The first coding element, the second coding element, the third coding element, the fourth coding element, the fifth coding element, the sixth coding element and the seventh coding element are used as first input data of the first data set; Using the non-coded data corresponding to the eighth coded element as the first data label of the first input data; taking the second coding element, the third coding element, the fourth coding element, the fifth coding element, the sixth coding element, the seventh coding element and the eighth coding element as second input data of the first data set; The non-coded data corresponding to the ninth coded element is used as the second data label of the first input data.

3. The user payment behavior prediction method based on trend, cause and confidence according to claim 1 or 2, characterized in that: The first loss function is a cross entropy loss function, and the first prediction model is established based on a Transformer model.

4. The user payment behavior prediction method based on trend, cause and confidence according to claim 1 or 2, characterized in that: The second coded data includes first non-payment data; the first non-payment data includes a first non-payment feature, a second non-payment feature, a third non-payment feature, a fourth non-payment feature, a fifth non-payment feature and a sixth non-payment feature; when performing the step of encoding the second data based on the non-payment feature to obtain the second coded data, the user payment behavior prediction method based on trend, reason and confidence also includes: Based on the first non-payment feature, a first feature value is obtained through a first preset condition; the first feature value includes a first upper limit value; Based on the second non-payment feature, a second feature value is obtained through a second preset condition; the second feature value includes a second upper limit value; Based on the third non-payment feature, a third feature value is obtained through a third preset condition; the third feature value includes a third upper limit value; Based on the fourth non-payment feature, a fourth feature value is obtained through a fourth preset condition; the fourth feature value includes a fourth upper limit value; Based on the fifth non-payment feature, a fifth feature value is obtained through a fifth preset condition; the fifth feature value includes a fifth upper limit value; Based on the sixth non-payment feature, a sixth feature value is obtained through a sixth preset condition; the sixth feature value includes a sixth upper limit value; Among them, the first upper limit value is smaller than the second upper limit value, the second upper limit value is smaller than the third upper limit value, the third upper limit value is smaller than the fourth upper limit value, the fourth upper limit value is smaller than the fifth upper limit value, and the fifth upper limit value is smaller than the sixth upper limit value.

5. The user payment behavior prediction method based on trend, cause and confidence according to claim 1 or 2, characterized in that: When executing the step of obtaining the payment prediction result according to the first combination probability, the second combination probability or the third combination probability, the user payment behavior prediction method based on trend, reason and confidence also includes: Confirm the first historical payment behavior and historical non-payment behavior of the user to be predicted; Inputting the first historical payment behavior into the first prediction model to obtain a first predicted payment rate; Inputting the historical non-payment behavior into the second prediction model to obtain a second predicted payment rate; Confirming a second historical payment behavior of the user to be predicted; Based on the second historical payment behavior, obtaining a payment rate to be tested; Based on the payment rate to be tested, determining a target payment level; the target payment level is the first paying user set, the second paying user set or the third paying user set; If the target payment level is the first paying user set, the payment prediction result is obtained according to the first combination probability.

6. The user payment behavior prediction method based on trend, cause and confidence according to claim 1 or 2, characterized in that: Before executing the step of dividing the user data into the first data and the second data, the method for predicting user payment behavior based on trend, reason and confidence also includes: The historical behavior log data of the user in the game platform is obtained, and the historical behavior log data is cleaned and preprocessed to obtain the user data.

7. The user payment behavior prediction method based on trend, cause and confidence according to claim 1 or 2, characterized in that: The second loss function is a cross entropy loss function, and the second prediction model is established based on a Transformer model.

8. The user payment behavior prediction method based on trend, cause and confidence according to claim 1 or 2, characterized in that: The encoding formula for calculating the first encoded data includes: ； in, Expressed as The day of the week corresponding to the day, the value is 1 to 7; Expressed as Users in the past Is there any payment behavior on that day? 9. A user payment behavior prediction system based on trends, causes and confidence, characterized in that: include: at least one processor; a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 8.

10. A computer readable medium having computer program instructions stored thereon, wherein the computer program instructions can be executed by a processor to implement the method according to any one of claims 1 to 8.

Images