Order placement probability prediction model training method and device, electronic equipment and chip

Abstract

This disclosure provides a training method, apparatus, electronic device, and chip for an order probability prediction model. The method includes: processing an original dataset using a first model to obtain a first dataset, the first dataset including change values ​​corresponding to at least one preset discount data; processing the original dataset using a second model to obtain a base order probability, the base order probability being independent of the preset discount data; constructing a second dataset based on the first dataset and the base order probability; and training a third model using the second dataset to obtain an order probability prediction model, which is used to predict the order probability corresponding to target discount data. This can improve the accuracy of model prediction.

Description

Technical Field

[0001] This disclosure relates to the field of data processing, and in particular to a method, apparatus, electronic device, chip, and storage medium for training an order probability prediction model. Background Technology

[0002] Using price discounts is a common promotional method. Predicting the probability of users placing orders under discounts can help merchants better configure product discounts and obtain the maximum profit at the lowest cost. The order probability is the ratio of the number of users who place orders for a certain product to the number of users who browse the product. When the order probability is not predicted accurately, it will lead to problems such as reduced profits and poor user shopping experience. Summary of the Invention

[0003] This disclosure provides a method, apparatus, electronic device, and chip for training an order probability prediction model to solve problems in related technologies.

[0004] The first aspect of this disclosure proposes a method for training an order probability prediction model. The method includes: processing an original dataset using a first model to obtain a first dataset, the first dataset including change values ​​corresponding to at least one preset discount data; processing the original dataset using a second model to obtain a basic order probability, the basic order probability being independent of the preset discount data; constructing a second dataset based on the first dataset and the basic order probability; and training a third model using the second dataset to obtain an order probability prediction model, the order probability prediction model being used to predict the order probability corresponding to target discount data.

[0005] In some embodiments of this disclosure, processing the original dataset using a first model to obtain the first dataset includes: collecting historical sales data and constructing an original dataset based on the historical sales data, the original dataset including at least one of at least one sample discount data, first probability data corresponding to each of the at least one sample discount data, and average sample discount data; training the first model using at least one sample discount data and the first probability data corresponding to each of the at least one sample discount data; determining second probability data corresponding to each of the at least one preset discount data using the trained first model, and determining the first dataset based on the second probability data corresponding to each of the at least one preset discount data.

[0006] In some embodiments of this disclosure, determining the first dataset based on the second probability data corresponding to each of the at least one preset discount data includes: determining the change value corresponding to each of the at least one preset discount data based on the average sample discount data and the second probability data corresponding to each of the at least one preset discount data; and determining the first dataset based on the change value corresponding to each of the at least one preset discount data.

[0007] In some embodiments of this disclosure, the process of using a second model to process the original dataset to obtain the basic order probability includes: removing at least one sample discount data from the original dataset to obtain a third dataset; and using the second model to make a prediction based on the third dataset to obtain the basic order probability.

[0008] In some embodiments of this disclosure, constructing a second dataset based on a first dataset and a basic order probability includes: summing the change values ​​corresponding to at least one preset discount data and the basic order probability to determine a third probability data corresponding to each of the at least one preset discount data; determining the third probability data as a fourth probability data corresponding to the preset discount data when the third probability data is greater than or equal to a first boundary value and less than a second boundary value; determining the first boundary value as the fourth probability data corresponding to the preset discount data when the third probability data is less than the first boundary value; determining the second boundary value as the fourth probability data corresponding to the preset discount data when the third probability data is greater than or equal to the second boundary value; and constructing a second dataset based on the fourth probability data corresponding to each of the at least one preset discount data.

[0009] In some embodiments of this disclosure, constructing a second dataset based on fourth probability data corresponding to at least one preset discount data includes: splitting the fourth probability data into first data and second data for each data in the fourth probability data corresponding to at least one preset discount data; determining the weight of the first data and the weight of the second data based on the fourth probability data; determining the label of the first data and the label of the second data; and constructing the second dataset based on at least one of the following: average sample discount data, at least one preset discount data, first data, second data, weight of the first data, weight of the second data, label of the first data, and label of the second data corresponding to at least one preset discount data.

[0010] In some embodiments of this disclosure, training a third model using a second dataset to obtain an order probability prediction model includes: constructing a third model based on at least one of average sample discount data and at least one preset discount data; and training the third model using the second dataset to obtain an order probability prediction model.

[0011] A second aspect of this disclosure provides a training apparatus for an order probability prediction model. The apparatus includes: a first processing unit for processing an original dataset using a first model to obtain a first dataset, the first dataset including change values ​​corresponding to at least one preset discount data; a second processing unit for processing the original dataset using a second model to obtain a basic order probability, the basic order probability being independent of the preset discount data; a third processing unit for constructing a second dataset based on the first dataset and the basic order probability; and a fourth processing unit for training a third model using the second dataset to obtain an order probability prediction model.

[0012] A third aspect of this disclosure provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the methods described in the first aspect of this disclosure.

[0013] A fourth aspect of this disclosure provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the methods described in the first aspect of this disclosure.

[0014] A fifth aspect of this disclosure provides a chip characterized by including at least one processor and a communication interface; the communication interface is used to receive signals input to the chip or signals output from the chip, and the processor communicates with the communication interface and implements the method described in the first aspect of this disclosure through logic circuits or executing code instructions.

[0015] A sixth aspect of this disclosure provides a computer program product including a computing program stored on a computer-readable storage medium. The computer program includes program instructions that, when executed by a computer, cause the computer to perform the method in any implementation of the first aspect described above.

[0016] In summary, the order probability prediction model training method proposed in this disclosure can process the original dataset using the first and second models to obtain training data for training the third model, thereby removing bias in the training data and improving the accuracy of model prediction.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0019] Figure 1 A flowchart illustrating a method for training an order probability prediction model according to an embodiment of this disclosure;

[0020] Figure 2 A flowchart illustrating a method for training an order probability prediction model according to an embodiment of this disclosure;

[0021] Figure 3 A flowchart illustrating a method for training an order probability prediction model according to an embodiment of this disclosure;

[0022] Figure 4A A flowchart illustrating a user order probability prediction method based on an overlay model provided in this embodiment of the disclosure;

[0023] Figure 4B A flowchart illustrating a data acquisition method provided in an embodiment of this disclosure;

[0024] Figure 4C A flowchart illustrating a data processing method provided in an embodiment of this disclosure;

[0025] Figure 4D A schematic flowchart of a model training method provided in an embodiment of this disclosure;

[0026] Figure 4E This is a schematic diagram of probability point changes provided in an embodiment of the present disclosure;

[0027] Figure 4F A flowchart illustrating a smoothing process provided in an embodiment of this disclosure;

[0028] Figure 5 This is a schematic diagram of the structure of a training device for an order probability prediction model provided in an embodiment of the present disclosure;

[0029] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure;

[0030] Figure 7 This is a schematic diagram of the chip structure provided in an embodiment of this disclosure. Detailed Implementation

[0031] Embodiments of this disclosure are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting this disclosure.

[0032] Online shopping is a very popular way of shopping, and in recent years, lifestyles such as buying groceries online, ordering electricity, and refueling online have gradually become accepted. For merchants providing online sales services, using price discounts is a common promotional method. Therefore, predicting the probability of users placing orders under discounts can help merchants better configure product discounts and obtain maximum profits with minimal cost.

[0033] Existing methods for predicting user order probability can be divided into two categories based on the data used:

[0034] 1. Directly use historical sales data, and use discounts as a feature to train a machine learning model to predict the probability of users placing orders.

[0035] 2. Add random experiments under different price discounts to make the discount data evenly distributed, and then use the discount as a feature to train a machine learning model to predict the probability of users placing orders.

[0036] However, the method described above, which directly uses historical sales data to train a machine learning model, ignores the bias in the original sales data. That is, there are certain factors that affect both the discount and the user's probability of placing an order. In this case, the results learned by the machine model may contradict common sense, such as the greater the discount, the lower the user's probability of placing an order.

[0037] Method 2 above uses random experiments with different price discounts. Although it solves the problem of uneven distribution of data sources, random experiments themselves need to cover a large number of users and take a long time, which requires high costs.

[0038] Therefore, in order to solve the above problems, this disclosure proposes a training method for an order probability prediction model. This method combines three models: LightGBM, Causal Forest DML, and Logistic Regression to predict the order probability of users under different discounts.

[0039] The specific details of this method are as follows.

[0040] Figure 1 This is a flowchart illustrating a method for training an order probability prediction model according to an embodiment of this disclosure. Figure 1 As shown, the method may include the following steps.

[0041] Step 101: Process the original dataset using the first model to obtain the first dataset.

[0042] In some embodiments, the first model may optionally be a model for eliminating data bias, such as a Casual Forest DML model, where data bias means that there are other influencing factors in addition to control factors. Therefore, the first model can be used to process the original dataset to eliminate data bias.

[0043] In some embodiments, the first dataset includes change values ​​corresponding to at least one preset discount data.

[0044] The preset discount is set according to actual needs. For example, when it is necessary to train the model to predict the probability of placing orders within a certain range, several discount values ​​can be determined from that range as preset discount data. For example, within the range [0.9, 1.0], the preset discount can be determined to include 0.9, 0.91, 0.92, ..., 0.99, 1.0, etc. That is, the preset discount can be determined according to the actual application scenario, and this disclosure does not restrict it.

[0045] In other words, because the coverage of discount data for each product is limited in actual sales scenarios, and for most products the discount values ​​are relatively concentrated and few in number (e.g., a certain product may only have two discounts: a larger discount of 0.90 and a smaller discount of 0.95), the original dataset can be processed using a first model to obtain a first dataset. This dataset can then yield the change value corresponding to at least one preset discount data point. This allows us to determine the impact of discounts on the probability of a user placing an order under various discount data points, where the change value represents the change in the order probability corresponding to the preset discount data point compared to the order probability corresponding to the average sample discount data point. In other words, by determining the change value corresponding to at least one preset discount data point, the coverage of discount data can be expanded, facilitating improved generalization ability of the model during training.

[0046] In some embodiments, processing the original dataset using a first model to obtain the first dataset includes: collecting historical sales data and constructing an original dataset based on the historical sales data, the original dataset including at least one of at least one sample discount data, first probability data corresponding to each of the at least one sample discount data, and average sample discount data; training the first model using at least one sample discount data and the first probability data corresponding to each of the at least one sample discount data; determining second probability data corresponding to each of the at least one preset discount data using the trained first model, and determining the first dataset based on the second probability data corresponding to each of the at least one preset discount data.

[0047] In some embodiments, optionally, when collecting historical sales data, at least one sample discount data and at least one first probability data corresponding to each sample discount data can be collected, wherein the first probability data is greater than or equal to 0 and less than or equal to 1, and the first probability data is the actual order probability of the corresponding user under the first sample discount data. That is, the actual product discount and the corresponding user order probability can be collected as training data.

[0048] In some embodiments, optionally, an average sample discount data can be determined for each product based on at least one sample discount data collected and the first probability data corresponding to each of the at least one sample discount data. That is, the average of all discount data for a product can be taken as the average sample discount data corresponding to that product.

[0049] In some embodiments, other feature data, such as weather information, date information, and product type, may also be collected during data collection, and this disclosure does not limit such data.

[0050] Step 102: The second model is used to process the original dataset to obtain the basic order probability.

[0051] In some embodiments, the second model may optionally be Light GBM, which can be used to estimate a base value of a user's order probability, i.e., the base order probability. The base order probability is independent of the preset discount data, i.e., the base order probability is not affected by the discount. For example, the user's order probability without considering the discount can be determined based on other features.

[0052] In some embodiments, processing the original dataset using a second model to obtain the basic order probability includes: removing at least one sample discount data from the original dataset to obtain a third dataset; and using the second model to make a prediction based on the third dataset to obtain the basic order probability.

[0053] Step 103: Construct the second dataset based on the first dataset and the basic order probability.

[0054] In some embodiments, the second dataset is used to train the third model to obtain a single probability prediction model. The second dataset includes at least one of the following: at least one preset discount data, average sample discount data, first data corresponding to at least one preset discount data, second data, weight of the first data, weight of the second data, label of the first data, and label of the second data.

[0055] In some embodiments, constructing a second dataset based on fourth probability data corresponding to at least one preset discount data includes: splitting the fourth probability data into first data and second data for each data in the fourth probability data corresponding to at least one preset discount data; determining the weight of the first data and the weight of the second data based on the fourth probability data; determining the label of the first data and the label of the second data; and constructing the second dataset based on at least one of the following: average sample discount data, at least one preset discount data, first data, second data, weight of the first data, weight of the second data, label of the first data, and label of the second data corresponding to at least one preset discount data.

[0056] Step 104: Use the second dataset to train the third model to obtain the order probability prediction model.

[0057] In some embodiments, the order probability prediction model is used to predict the order probability corresponding to the target discount data. This means that the order probability prediction model can predict the order probability of users corresponding to different discount data, making it easier to determine the optimal discount data and obtain the maximum profit at the lowest cost.

[0058] In some embodiments, training a third model using a second dataset to obtain an order probability prediction model includes: constructing a third model based on at least one of average sample discount data and at least one preset discount data; and training the third model using the second dataset to obtain an order probability prediction model.

[0059] In summary, the above embodiments of this application can process the original dataset using the first and second models to obtain training data for training the third model, thereby removing bias in the training data and improving the accuracy of model predictions.

[0060] Figure 2 This is a flowchart illustrating a method for training an order probability prediction model according to an embodiment of this disclosure. Figure 2 As shown, based on Figure 1 The illustrated embodiment shows that the method includes the following steps.

[0061] Step 201: Sum the change value and basic order probability corresponding to each of the at least one preset discount data to determine the third probability data corresponding to each of the at least one preset discount data.

[0062] In some embodiments, a first model can be used to process the original dataset to obtain a first dataset. The first dataset includes change values ​​corresponding to at least one preset discount data. The method includes: collecting historical sales data and constructing an original dataset based on the historical sales data. The original dataset includes at least one of at least one sample discount data, first probability data corresponding to each of the at least one sample discount data, and average sample discount data; training the first model using the at least one sample discount data and the first probability data corresponding to each of the at least one sample discount data; determining second probability data corresponding to each of the at least one preset discount data using the trained first model, and determining the first dataset based on the second probability data corresponding to each of the at least one preset discount data.

[0063] In some embodiments, at least one sample discount data may be used. K The first model is trained using the first probability data (CVR) corresponding to at least one sample discount data, where the discount data can be used to train the first model. K Input the first model and use the first model to predict the discount. K The corresponding order probability, and based on the estimated discount. K The parameters of the model are adjusted based on the difference between the corresponding order probability and the first probability data to obtain the trained first model.

[0064] In some embodiments, the original dataset may optionally include a training dataset and a test dataset. That is, after training the first model, the first model can be tested using the test dataset, and the parameters of the first model can be adjusted according to the test results to obtain the trained first model, which can improve the prediction accuracy and generalization ability of the model.

[0065] In other words, after collecting historical sales data, training and test datasets can be constructed based on the historical sales data. Optionally, the historical sales data can be processed to generate training and test datasets based on the processed data. Data processing may include, for example, data cleaning, anomaly handling, data feature extraction, etc. This disclosure does not limit the scope of the data processing.

[0066] In some embodiments, after obtaining the trained first model, the trained first model can be used to determine the second probability data corresponding to at least one preset discount data in the original dataset, and the first dataset can be determined based on the second probability data corresponding to each of the at least one preset discount data. In other words, the trained first model can be used to predict the second probability data corresponding to at least one preset discount data, that is, it can predict the probability of a user placing an order under the preset discount.

[0067] In some embodiments, determining the first dataset based on the second probability data corresponding to each of the at least one preset discount data includes: determining the change value corresponding to each of the at least one preset discount data based on the average sample discount data and the second probability data corresponding to each of the at least one preset discount data; and determining the first dataset based on the change value corresponding to each of the at least one preset discount data.

[0068] In some embodiments, the first dataset may include at least one preset discount data corresponding to a change value delta. i The change value is the change in user probability under the second probability data relative to the mean-discount data. That is, the user order probability corresponding to the mean-discount data can be determined based on the data in the original dataset. In some embodiments, optionally, when there is no user order probability corresponding to the mean-discount data in the original dataset, the trained first model can be used to determine the user order probability corresponding to the mean-discount data. In this case, based on the user order probability corresponding to the mean-discount data and the second probability data, a change value can be determined for each preset discount data. For example, the difference between the user order probability corresponding to the second probability data and the mean-discount data can be determined as the change value corresponding to the preset discount data.

[0069] In some embodiments, the first dataset may optionally include, in addition to the change values ​​corresponding to each of the at least one preset discount data, average discount data, other feature data, etc., which are not limited in this disclosure.

[0070] In some embodiments, processing the original dataset using a second model to obtain the basic order probability includes: removing at least one sample discount data from the original dataset to obtain a third dataset; and using the second model to make a prediction based on the third dataset to obtain the basic order probability.

[0071] In some embodiments, the base order probability is the base order probability unaffected by discounts. That is, the sample discount data in the original dataset can be removed, and the remaining data can constitute the third dataset mentioned above. The second model can use the third dataset to determine a base order probability base_cvr, which is unrelated to the sample discount data. Optionally, a base order probability can be determined for a product. For example, the user order probability of a certain product when it is unaffected by discounts can be determined, or the analysis can be performed on a specific date to determine the corresponding base order probability. That is, the user order probability when it is unaffected by discounts at different times can be determined, etc. This disclosure does not limit this.

[0072] In some embodiments, the change values ​​corresponding to at least one preset discount data and the basic order probability can be summed to obtain the third probability data corresponding to each of the at least one preset discount data. This can update the user order probability corresponding to at least one preset discount data, further reduce the impact of other factors on the user order probability, and more accurately reflect the impact of discounts on the user order probability.

[0073] Step 202: When the third probability data is greater than or equal to the first boundary value and less than the second boundary value, the third probability data is determined to be the fourth probability data corresponding to the preset discount data.

[0074] In some embodiments, after updating the user order probability corresponding to the preset discount data, in order to avoid abnormal probability data, it is necessary to limit the range of the third probability data. Optionally, the value of the third probability is greater than or equal to 0 and less than or equal to 1, that is, the range is [0,1].

[0075] In some embodiments, the first boundary value is the lower boundary value, and the second boundary value is the upper boundary value. When the range is [0,1], the first boundary value is 0, and the second boundary value is 1. When the third probability data is within the range, it is not necessary to update the third probability data; the third probability data can be directly used as the fourth probability data corresponding to the preset discount data. The fourth probability data is the order probability corresponding to the updated preset discount data. That is, when the third probability data is greater than or equal to the first boundary value and less than the second boundary value, the third probability data is determined to be the fourth probability data corresponding to the preset discount data.

[0076] Step 203: When the third probability data is less than the first boundary value, the first boundary value is determined to be the fourth probability data corresponding to the preset discount data.

[0077] In some embodiments, when the third probability data is not within the range, the fourth probability data corresponding to the preset discount data can be determined as the first boundary value or the second boundary value. Specifically, when the third probability data is less than the first boundary value, the first boundary value is determined as the fourth probability data corresponding to the preset discount data.

[0078] Step 204: When the third probability data is greater than or equal to the second boundary value, determine the second boundary value as the fourth probability data corresponding to the preset discount data.

[0079] In some embodiments, when the third probability data is outside the range, the fourth probability data corresponding to the preset discount data can be determined as either the first boundary value or the second boundary value. Specifically, when the third probability data is greater than or equal to the second boundary value, the second boundary value is determined to be the fourth probability data cvr_pred corresponding to the preset discount data. i .

[0080] In summary, by limiting the range of the third probability data, the fourth probability is obtained, as shown in Formula 1 below:

[0081] cvr_pred i =max(min(base_cvr+delta) i Formula 1 (i = 0, 1, 2, ..., N)

[0082] Among them, cvr_pred i This is the fourth probability data corresponding to the i-th preset discount data, where base_cvr is the base order probability, and delta is the delta. i This represents the change value corresponding to the i-th preset discount data.

[0083] Step 205: Construct a second dataset based on the fourth probability data corresponding to at least one preset discount data.

[0084] In some embodiments, the second dataset can be used to train the third model to obtain an order probability prediction model, wherein the third model can be a logistic regression model.

[0085] In some embodiments, constructing a second dataset based on fourth probability data corresponding to at least one preset discount data includes: splitting the fourth probability data into first data and second data for each data in the fourth probability data corresponding to at least one preset discount data; determining the weight of the first data and the weight of the second data based on the fourth probability data; determining the label of the first data and the label of the second data; and constructing the second dataset based on at least one of the following: average sample discount data, at least one preset discount data, first data, second data, weight of the first data, weight of the data, label of the first data, and label of the second data corresponding to at least one preset discount data.

[0086] In other words, the fourth probability data can be split to obtain two data with different labels, that is, the first data and the second data have different labels. The labels can be binary labels, that is, the first data and the second data can be distinguished by 0 and 1. In other words, the label of the first data can be 0 or 1, and the label of the second data can be 0 or 1.

[0087] In some embodiments, the fourth probability data can be split into a data with a probability of 1 and a data with a probability of 0. That is, the probability values ​​of the first data and the second data can be the labels of the first data and the second data, for example, in the fourth probability data cvr_pred i When the value is 0.15, it can be split into the first data cvr_pred i =0 and the second data cvr_pred i=1. After determining the first and second data, the weights of the first and second data can be determined based on the fourth probability data. The value of the fourth probability data is the sum of the first data multiplied by its weight and the second data multiplied by its weight. For example, in the fourth probability data cvr_pred i When the value is 0.15, the weight of the first data is 85, and the weight of the second data is 15.

[0088] In other words, the fourth probability data can be replaced with the first and second data, that is, the fourth probability data can be split into two parts, namely the first data and the second data, which facilitates the training of the third model. Specifically, the formula for calculating the weight of the first data, weight0, is as follows:

[0089] weight0 = 100 * (1 - cvr_pred) i ) Formula 2

[0090] The formula for calculating the weight1 of the second data is as follows: Formula 3:

[0091] weight1 = 100 * cvr_pred i Formula 3

[0092] cvr_pred i This represents the value of the fourth probability data.

[0093] In some embodiments, the second dataset may also include other feature data, such as weather information, date information, and types of goods, etc., which are not limited in this disclosure.

[0094] In summary, the above embodiments of this application can update the user order probability data corresponding to the preset discount data according to the change value and the basic order probability, and split the updated probability data to obtain a first data and a second data, as well as the labels and weights corresponding to the first data and the second data respectively. A second dataset can be constructed, which is convenient for training the third model based on the second dataset. It can realize the use of unbiased data to train the model and improve the accuracy of model prediction.

[0095] Figure 3 This is a flowchart illustrating a method for training an order probability prediction model according to an embodiment of this disclosure. Figure 3 As shown, based on Figure 1 The illustrated embodiment shows that the method includes the following steps.

[0096] Step 301: Construct a third model based on at least one of the average sample discount data and at least one preset discount data.

[0097] In some embodiments, the third model can be a logistic regression model, and the third model is constructed based on at least one of the average sample discount data and at least one preset discount data, as shown in Formula 4 below:

[0098] constrained_cvr_pred i

[0099] =Sigmoid(coef0 × discount) i +coef1×mean_discount+coef2×other_feat

[0100] +debias),coef0≤0,i=0,1,2,…,N Formula 4

[0101] Among them, consttrained_cvr_pred i For the third model, in discount data i The predicted order probability is obtained from the user's prediction. mean_discount is the average sample discount data, other_feat is other feature data, that is, other feature data can be used to build a third model, and debias is the bias value used to correct the bias of the third model.

[0102] In some embodiments, although the logistic regression model is a classification model, in this disclosure, the sigmoid function can be used to obtain the final predicted order probability value, or probability score, rather than the 0,1 value after classification, such as the constrained_cvr_pred value. i It can be 0.95, etc.

[0103] Step 302: Train the third model using the second dataset to obtain the order probability prediction model.

[0104] In some embodiments, the third model can be trained using the second dataset to obtain an order probability prediction model. Optionally, at least one preset discount data in the second dataset can be used as input to the model to predict the order probability corresponding to at least one preset discount data. The predicted order probability is then compared with the fourth probability data corresponding to at least one preset discount data in the second dataset. The third model is adjusted based on the comparison results to achieve training of the third model.

[0105] In some embodiments, the second dataset may optionally include a training dataset and a test dataset. That is, after training the third model, the third model can be tested using the test dataset, and the parameters of the third model can be adjusted according to the test results to obtain an order probability prediction model, which can improve the accuracy and generalization ability of the model's prediction.

[0106] In some embodiments, after training the third model to obtain the order probability prediction model, the relationship between discount and order probability can be obtained with other variables remaining unchanged. Optionally, the relationship between discount and order probability can be displayed by fitting a curve, which can be represented by the following formula 5:

[0107] constrained_cvr_pred i =f(discount) i ), i = 0, 1, 2, ..., N Formula 5

[0108] Optionally, the curve is displayed as a smooth and monotonic curve, with monotonicity either decreasing or remaining constant. That is, the larger the value of the discount data (closer to 1), the more expensive the corresponding product price and the lower the probability of placing an order. Conversely, the smaller the value of the discount data (closer to 0), the cheaper the corresponding product price and the higher the probability of placing an order.

[0109] In summary, the above embodiments of this disclosure can use data to eliminate bias to train the third model, thereby obtaining an order probability prediction model and improving the accuracy of the model's predictions.

[0110] The technical solutions of this disclosure will be further described in detail below with reference to specific application embodiments.

[0111] The following is a user order probability prediction method based on a superposition model provided in this disclosure. This method combines three models—LightGBM, CausalForestDML (CausalForestDML), and logistic regression—to predict the order probability of users under different discounts. CausalForestDML combines DML (Dual Machine Learning) and CausalForest, which can better estimate the heterogeneity treatment effect in causal inference problems, i.e., estimate the response of different individuals or subgroups to different treatment strategies in high-dimensional and complex nonlinear relationships. CausalForest is a model composed of multiple causal trees. Each causal tree divides the original data samples into a control group and a control group based on whether a strategy is applied. By maximizing the difference in the probability distribution of whether an order is placed between the control group and the control group during splitting, the distribution of features other than strategy features in the samples of each leaf node is made as similar as possible, thereby eliminating data bias.

[0112] This solution first uses LightGBM to estimate a baseline value for the probability of a user placing an order. Then, it uses the CausalForestDML model to estimate the change in the probability of a user placing an order under each discount compared to the baseline value. The baseline value and the change are added together to calculate the probability of a user placing an order under each discount. Finally, these predicted probability points are split into positive and negative samples to train a logistic regression model. The monotonicity of the coefficients in the logistic regression model is constrained, and the probability predicted by the logistic regression is used as the final probability of a user placing an order.

[0113] like Figure 4A As shown, the user order probability prediction method proposed in this scheme consists of three parts: sample generation, probability prediction, and probability constraint. The user order probability prediction system implementing this scheme includes a sample generation unit, a superimposed model probability prediction unit, and a smooth and monotonic probability constraint unit. The user order probability prediction method is described in detail below.

[0114] 1.1 The sample generation unit uses historical sales data to construct training and test samples, such as... Figure 4B The process of collecting samples for the above sample generation unit includes:

[0115] 1) In the sample, CVR represents the probability of a user placing an order, and discount represents the discount of the product. Because the selling prices of different products vary greatly, discount is used as the feature of the sample, 1≥discount≥0. Possible discount values ​​are such as 0.90, 0.91, 0.92, ..., 0.99, 1.0, etc. In historical sales data, the discount coverage of each product is not complete, and most of them are concentrated around a few discounts. For example, a certain product only has two discounts, a larger discount of 0.90 and a smaller discount of 0.95.

[0116] 2) The mean-discount in the sample represents the historical average discount data of the product. Optionally, a historical average discount data can be determined for a single product.

[0117] 3) Other features indicate other characteristics such as weather, date, type of goods, etc.

[0118] 1.2 The probability prediction unit of the superimposed model uses Causal Forest-Dual Machine Learning (hereinafter referred to as CasualForestDM L) and LightGBM to make unbiased predictions of the probability of users placing orders, such as... Figure 4C The specific process of the probability prediction method for the superposition model includes:

[0119] 1) Input the samples obtained in step 1 into CasualForestDML for training, where discount KAs a strategy, CVR is used as the label for training CasualForestDML. During prediction, mean-discount is used as the base discount value, while discount is also used. i As a reference discount value, discount i Given a preset discount value, the predicted result is delta. i Indicates the use of discount i As a value representing the change in the probability of placing an order compared to the mean discount, CasualForestDML will apply this value to all discounts. i Delta is obtained by predicting all values ​​of i = 0, 1, 2, ..., N. i , i = 0, 1, 2, ..., N.

[0120] 2) Remove the discount from the samples obtained in step 1. i The data is fed into LightGBM for training. The predicted result, base_cvr, represents the base order probability when the user is not affected by price discounts. Then, base_cvr is compared with delta. i Add them together and restrict the result to the range [0.0, 1.0] to get cvr_pred i :

[0121] cvr_pred i =max(min(base_cvr+delta) i ,1.0),0.0),i=0,1,2,…,N

[0122] cvr_pred i This represents the predicted probability of a user placing an order under different discounts.

[0123] 1.3 The smooth and monotonic probability constraint unit uses logistic regression to apply monotonicity and smoothness constraints to the predicted user order probabilities in step 2. For example... Figure 4D The specific process includes:

[0124] 1) Combine the CVR and discount of the sample obtained in step 1 above. K Replace them with cvr_pred obtained using the two methods described above. i and discount i , i = 0, 1, 2, ..., N, to obtain N+1 samples.

[0125] 2) Sample replication and weight configuration methods: The logistic regression algorithm requires samples to have binary labels of 0 or 1. Therefore, the samples obtained in step 2 above need to be processed. For example, for a cvr_pred iA sample with a value of 0.15 needs to be split into a single cvr_pred. i =0 and cvr_pred i For samples with a value of 1, for cvr_pred i The sample with a value of 0 is assigned a weight of 85, cvr_pred i The sample with a value of 1 is assigned a weight of 15, and the weight formula is as follows:

[0126] weight0 = 100 * (1 - cvr_pred) i )

[0127] weight1 = 100 * cvr_pred i

[0128] 3) Logistic Regression-Based Smooth Monotonic Order Probability Prediction Method: The samples from 1) are fed into a logistic regression dataset for training, where the discount... i The corresponding coefficients (coef0 in the following formula) are restricted to being non-positive, and the weights during logistic regression training are configured as in b). The prediction formula for logistic regression can be expressed as follows:

[0129] constrained_cvr_pred i

[0130] =Sigmoid(coef0 × discount) i +coef1×mean_discount+coef2×other_feat

[0131] +debias),coef0≤0,i=0,1,2,…,N

[0132] Note that although logistic regression is a classification model, the final prediction result here is constrained_cvr_pred i The values ​​i = 0, 1, 2, ..., N represent the probability scores directly predicted by the logistic regression algorithm, not the 0 or 1 values ​​after classification. The curve shown in the following formula represents the relationship between the discount and the probability of placing an order, assuming other variables remain constant.

[0133] constrained_cvr_pred i =f(discount) i ), i = 0, 1, 2, ..., N

[0134] It is a smooth and monotonic curve, with monotonicity either decreasing or remaining constant.

[0135] The above method will be explained and illustrated through a specific embodiment below.

[0136] 2.1 First, construct a sample set D from the original sales data, and input D into CasualForestDML and LightGBM to train and obtain models Model0 and Model1.

[0137] For a specific input test sample S 0 Its user order probability and specific discount characteristics are as follows:

[0138] cvr 0 =0.15

[0139] discount 0 =0.95

[0140] mean_discount 0 =0.97

[0141] Predict S0 using Model0, with 0.97 as the base discount value.

[0142] discount i =[0.90,0.91,0.92,..,0.99,1.0]

[0143] As the discount value to be applied, the change in order probability relative to the average discount is predicted for each discount value, resulting in delta as shown in Table 1 below:

[0144] discount 0.90 0.91 0.92 0.93 0.94 0.95 0.96 0.97 0.98 0.99 1.00 delta 0.077 0.051 0.052 0.035 0.021 0.026 0.013 -.010 -0.017 -0.012 -0.040

[0145] Use Model1 to compare with S 0 The prediction yielded the following:

[0146] base_cvr = 0.12

[0147] According to the formula cvr_pred i =max(min(base_cvr+delta) i The predicted cvr_pred values ​​for i = 0, 1, 2, ..., N are shown in Table 2 below, and the corresponding probability point changes are as follows: Figure 4E As shown:

[0148] discount 0.90 0.91 0.92 0.93 0.94 0.95 0.96 0.97 0.98 0.99 1.00 cvr_pred 0.197 0.171 0.172 0.155 0.141 0.146 0.133 0.110 0.103 0.108 0.080

[0149] 2.2 Split each order probability point predicted in step 1 above into two samples according to the method in step 1.3, as shown in Table 3 below. Use new_cvr_pred in Table 3 as the label, discount and mean_discount as features, and weight as the sample weight. Train the input logistic regression algorithm according to the method in step 1.3 to obtain model Model3.

[0150] discount mean_discount cvr_pred weight new_cvr_pred 0.90 0.97 0.197 80.3 0 0.91 0.97 0.171 82.9 0 0.92 0.97 0.172 82.8 0 0.93 0.97 0.155 84.5 0 0.94 0.97 0.141 85.9 0 0.95 0.97 0.146 85.4 0 0.96 0.97 0.133 86.7 0 0.97 0.97 0.11 89 0 0.98 0.97 0.103 89.7 0 0.99 0.97 0.108 89.2 0 1.00 0.97 0.08 92 0 0.90 0.97 0.197 19.7 1 0.91 0.97 0.171 17.1 1 0.92 0.97 0.172 17.2 1 0.93 0.97 0.155 15.5 1 0.94 0.97 0.141 14.1 1 0.95 0.97 0.146 14.6 1 0.96 0.97 0.133 13.3 1 0.97 0.97 0.11 11 1 0.98 0.97 0.103 10.3 1 0.99 0.97 0.108 10.8 1 1.00 0.97 0.08 8 1

[0151] Modify the discount value of sample S0 by setting discount = [0.90, 0.91, 0.92, ..., 0.99, 1.0], keeping other features unchanged, and predict the smooth and monotonic order probability cvr_constrained_pred for users under different discounts, such as... Figure 4F The blue curve in the image.

[0152] In summary, the above examples disclosed herein can be combined with LightGBM and CausalForestDML to process the collected dataset, which can eliminate the bias of the data. The logistic regression model can be trained based on the processed data, and the trained model can be used to predict the probability of users placing orders under different discounts, thereby improving the model's prediction probability.

[0153] Figure 5 This is a schematic diagram of the structure of an order probability prediction model training device 500 provided in an embodiment of this disclosure. Figure 5 As shown, the device includes: a first processing unit 510, used to process the original dataset using a first model to obtain a first dataset, the first dataset including the change values ​​corresponding to at least one preset discount data; a second processing unit 520, used to process the original dataset using a second model to obtain a basic order probability, the basic order probability being independent of the preset discount data; a third processing unit 530, used to construct a second dataset based on the first dataset and the basic order probability; and a fourth processing unit 540, used to train the third model using the second dataset to obtain an order probability prediction model.

[0154] In some embodiments, the first processing unit is further configured to collect historical sales data and construct an original dataset based on the historical sales data. The original dataset includes at least one of at least sample discount data, first probability data corresponding to each of the at least sample discount data, and average sample discount data. The first model is trained using at least one sample discount data and the first probability data corresponding to each of the at least one sample discount data. The trained first model is used to determine second probability data corresponding to each of the at least one preset discount data, and the first dataset is determined based on the second probability data corresponding to each of the at least one preset discount data.

[0155] In some embodiments, the first processing unit is further configured to determine the change value corresponding to each of the at least one preset discount data based on the average sample discount data and the second probability data corresponding to each of the at least one preset discount data; and to determine the first dataset based on the change value corresponding to each of the at least one preset discount data.

[0156] In some embodiments, the second processing unit is further configured to remove at least one sample discount data from the original dataset to obtain a third dataset; and to use the second model to make a prediction based on the third dataset to obtain the basic order probability.

[0157] In some embodiments, the third processing unit is further configured to sum the change values ​​and basic order probabilities corresponding to each of the at least one preset discount data to determine the third probability data corresponding to each of the at least one preset discount data; when the third probability data is greater than or equal to a first boundary value and less than a second boundary value, determine the third probability data as the fourth probability data corresponding to the preset discount data; when the third probability data is less than the first boundary value, determine the first boundary value as the fourth probability data corresponding to the preset discount data; when the third probability data is greater than or equal to the second boundary value, determine the second boundary value as the fourth probability data corresponding to the preset discount data; and construct a second dataset based on the fourth probability data corresponding to each of the at least one preset discount data.

[0158] In some embodiments, the third processing unit is further configured to split the fourth probability data into first data and second data for each data in the fourth probability data corresponding to at least one preset discount data; determine the weight of the first data and the weight of the second data according to the fourth probability data; determine the label of the first data and the label of the second data; and construct a second dataset according to at least one of the average sample discount data, at least one preset discount data, the first data, the second data, the weight of the first data, the weight of the second data, the label of the first data, and the label of the second data corresponding to each of the at least one preset discount data.

[0159] In some embodiments, the fourth processing unit is further configured to construct a third model based on at least one of the average sample discount data and at least one preset discount data; and to train the third model using the second dataset to obtain an order probability prediction model.

[0160] In summary, the order probability prediction model training device 500 can process the original dataset using the first and second models to obtain training data for training the third model, thereby removing bias in the training data and improving the accuracy of model prediction.

[0161] The methods and apparatus provided in the embodiments of this application have been described above. To implement the functions of the methods provided in the embodiments of this application, the electronic device may include a hardware structure and software modules, and may implement the above functions in the form of a hardware structure, software modules, or a hardware structure plus software modules. One of the above functions may be executed in the form of a hardware structure, software modules, or a hardware structure plus software modules.

[0162] Figure 6 This is a block diagram illustrating an electronic device 600 for implementing the above-described method according to an exemplary embodiment. For example, the electronic device 600 may be a mobile phone, computer, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0163] Reference Figure 6 The electronic device 600 may include one or more of the following components: a processing component 602, a memory 604, a power supply component 606, a multimedia component 608, an audio component 610, an input / output (I / O) interface 612, a sensor component 614, and a communication component 616.

[0164] Processing component 602 typically controls the overall operation of electronic device 600, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 602 may include one or more processors 620 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 602 may include one or more modules to facilitate interaction between processing component 602 and other components. For example, processing component 602 may include a multimedia module to facilitate interaction between multimedia component 608 and processing component 602.

[0165] Memory 604 is configured to store various types of data to support the operation of electronic device 600. Examples of this data include instructions for any application or method operating on electronic device 600, contact data, phonebook data, messages, pictures, videos, etc. Memory 604 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0166] Power supply component 606 provides power to various components of electronic device 600. Power supply component 606 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 600.

[0167] Multimedia component 608 includes a screen that provides an output interface between electronic device 600 and user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 608 includes a front-facing camera and / or a rear-facing camera. When electronic device 600 is in an operating mode, such as a shooting mode or video mode, the front-facing camera and / or rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0168] Audio component 610 is configured to output and / or input audio signals. For example, audio component 610 includes a microphone (MIC) configured to receive external audio signals when electronic device 600 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 604 or transmitted via communication component 616. In some embodiments, audio component 610 also includes a speaker for outputting audio signals.

[0169] I / O interface 612 provides an interface between processing component 602 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0170] Sensor assembly 614 includes one or more sensors for providing state assessments of various aspects of electronic device 600. For example, sensor assembly 614 may detect the on / off state of electronic device 600, the relative positioning of components such as the display and keypad of electronic device 600, changes in position of electronic device 600 or a component of electronic device 600, the presence or absence of user contact with electronic device 600, orientation or acceleration / deceleration of electronic device 600, and temperature changes of electronic device 600. Sensor assembly 614 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 614 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 614 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0171] Communication component 616 is configured to facilitate wired or wireless communication between electronic device 600 and other devices. Electronic device 600 can access wireless networks based on communication standards, such as WiFi, 2G or 3G, 4G LTE, 5G NR (NewRadio), or combinations thereof. In one exemplary embodiment, communication component 616 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 616 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0172] In an exemplary embodiment, the electronic device 600 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.

[0173] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 604 including instructions, which can be executed by a processor 620 of an electronic device 600 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0174] Embodiments of this disclosure also provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the methods described in the above embodiments of this disclosure.

[0175] Embodiments of this disclosure also provide a computer program product, including a computer program that is executed by a processor using the methods described in the above embodiments of this disclosure.

[0176] Figure 7 This is a schematic diagram illustrating the structure of a chip 700 for implementing the above method according to an exemplary embodiment. (Refer to...) Figure 7 The chip 700 includes a communication interface 701 and at least one processor 702. The communication interface 701 is used to receive signals input to the chip 700 or signals output from the chip 700. The processor 702 communicates with the communication interface 701 and implements the methods described in the above embodiments of this disclosure through logic circuits or executing code instructions.

[0177] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0178] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in at least one embodiment or example.

[0179] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0180] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processing module, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having at least one wiring (control method), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic device, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0181] It should be understood that various parts of the embodiments of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0182] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0183] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc.

[0184] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for training an order probability prediction model, characterized in that, The method includes: The original dataset is processed using the first model to obtain the first dataset, which includes the change value corresponding to at least one preset discount data. The original dataset is processed using a second model to obtain a basic order probability, which is independent of the preset discount data. Based on the first dataset and the basic order probability, a second dataset is constructed; The third model is trained using the second dataset to obtain an order probability prediction model, which is used to predict the order probability corresponding to the target discount data.

2. The method according to claim 1, characterized in that, The first dataset obtained by processing the original dataset using the first model includes: Collect historical sales data and construct the original dataset based on the historical sales data. The original dataset includes at least one of the following: at least one sample discount data, first probability data corresponding to each of the at least one sample discount data, and average sample discount data. The first model is trained using the at least one sample discount data and the first probability data corresponding to each of the at least one sample discount data; The trained first model is used to determine the second probability data corresponding to each of the at least one preset discount data, and the first dataset is determined based on the second probability data corresponding to each of the at least one preset discount data.

3. The method according to claim 2, characterized in that, Determining the first dataset based on the second probability data corresponding to each of the at least one preset discount data includes: Based on the average sample discount data and the second probability data corresponding to each of the at least one preset discount data, determine the change value corresponding to each of the at least one preset discount data; The first dataset is determined based on the change value corresponding to each of the at least one preset discount data.

4. The method according to claim 1, characterized in that, The second model is used to process the original dataset to obtain the basic order probability, which includes: By removing at least one discounted sample from the original dataset, a third dataset is obtained. Based on the third dataset, the second model is used to make predictions to obtain the basic order probability.

5. The method according to claim 3, characterized in that, The step of constructing the second dataset based on the first dataset and the basic order probability includes: The change values ​​corresponding to each of the at least one preset discount data and the basic order probability are summed to determine the third probability data corresponding to each of the at least one preset discount data. When the third probability data is greater than or equal to the first boundary value and less than the second boundary value, the third probability data is determined to be the fourth probability data corresponding to the preset discount data; When the third probability data is less than the first boundary value, the first boundary value is determined to be the fourth probability data corresponding to the preset discount data; When the third probability data is greater than or equal to the second boundary value, the second boundary value is determined to be the fourth probability data corresponding to the preset discount data; The second dataset is constructed based on the fourth probability data corresponding to each of the at least one preset discount data.

6. The method according to claim 5, characterized in that, The step of constructing the second dataset based on the fourth probability data corresponding to each of the at least one preset discount data includes: For each piece of data in the fourth probability data corresponding to the at least one preset discount data, the fourth probability data is split into first data and second data; The weights of the first data and the second data are determined based on the fourth probability data. Determine the labels of the first data and the second data; The second dataset is constructed based on the average sample discount data, the at least one preset discount data, and at least one of the first data, the second data, the weight of the first data, the weight of the second data, the label of the first data, and the label of the second data corresponding to each of the at least one preset discount data.

7. The method according to claim 2, characterized in that, The step of training the third model using the second dataset to obtain the order probability prediction model includes: The third model is constructed based on at least one of the average sample discount data and at least one preset discount data. The third model is trained using the second dataset to obtain the order probability prediction model.

8. A training device for an order probability prediction model, characterized in that, The device includes: The first processing unit is used to process the original dataset using a first model to obtain a first dataset, wherein the first dataset includes at least one preset discount data corresponding to a change value. The second processing unit is used to process the original dataset using a second model to obtain a basic order probability, wherein the basic order probability is independent of the preset discount data. The third processing unit is used to construct a second dataset based on the first dataset and the basic order probability; The fourth processing unit is used to train the third model using the second dataset to obtain an order probability prediction model.

9. An electronic device, characterized in that, include: At least one processor; as well as 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 to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

11. A chip, characterized in that, It includes at least one processor and a communication interface; the communication interface is used to receive signals input to the chip or signals output from the chip, and the processor communicates with the communication interface and implements the method as described in any one of claims 1 to 7 through logic circuits or executing code instructions.

12. A computer program product comprising a computer program, wherein, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images