Data processing method, model training method, recommendation method, equipment and product

Abstract

The invention discloses a data processing method, a model training method, a recommendation method, equipment and a product, and the method comprises the steps: determining the feature data of a first user and a second user, the feature data comprises the user features of the user to which the feature data belongs, a first moment when a preset interaction event is carried out with the user, and an interaction result corresponding to the preset interaction event; then, first feature data and second feature data corresponding to a preset moment are determined from the feature data, the first moment in the first feature data is equal to the preset moment, and the first moment in the second feature data is different from the preset moment; finally, respectively identifying the first feature data and the second feature data to obtain a first sample belonging to the experimental group and a second sample belonging to the control group at a preset moment; according to the invention, samples with higher effectiveness can be generated for the gain model.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and specifically to a data processing method, a model training method, a recommendation method, a device, and a product. Background Art

[0002] Currently, in business, the results of user interactions are influenced by the timing of those interactions. For example, in common recommendation scenarios, to foster deeper communication with users and promote successful recommendations, it's often necessary to consider personalized recommendation timing to improve both effectiveness and success rate. Uplift models are often used to evaluate the impact of certain interventions (such as specific interaction timing) on ​​user behavior.

[0003] However, in related technologies, time is generally used as one of the features in sample data to train the gain model, so that the model estimates the recommendation success rate of users at different times, obtains the success rate gain between different times, and then recommends time based on the gain.

[0004] However, with this type of sample data, the model can only learn the success rate gain between different fixed moments. Summary of the Invention

[0005] In response to the above technical problems, the embodiments of the present application provide a data processing method, a model training method, a recommendation method, a device and a product, which can generate more effective samples for the gain model.

[0006] In a first aspect, an embodiment of the present application provides a data processing method, comprising:

[0007] Determining characteristic data of a first user and a second user, wherein the characteristic data includes user characteristics of the user to which the characteristic data belongs, a first time of a preset interaction event with the user, and an interaction result corresponding to the preset interaction event; the first time corresponding to the first user is a time when the preset interaction event is recommended for the first user; the first time corresponding to the second user is different from the time when the preset interaction event is recommended for the second user;

[0008] Determining first characteristic data and second characteristic data corresponding to a preset time from the characteristic data, wherein the first time in the first characteristic data is equal to the preset time, and the first time in the second characteristic data is different from the preset time;

[0009] The first characteristic data and the second characteristic data are respectively marked to obtain a first sample belonging to the experimental group and a second sample belonging to the control group at a preset moment, which are used as sample data of the gain model.

[0010] In a second aspect, an embodiment of the present application provides a model training method, comprising:

[0011] Obtaining sample data of the gain model to be trained, wherein the sample data includes a first sample belonging to the experimental group and a second sample belonging to the control group at at least one preset time determined by the data processing method of the present application;

[0012] Based on the first sample and the second sample, the gain model is used to predict a first probability of users in the first sample and a second probability of users in the second sample, respectively, where the first probability is the probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a preset time, and the second probability is the probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a random time other than the preset time;

[0013] Calculating the loss of the gain model based on the interaction result and the first probability in the first sample, and the interaction result and the second probability in the second sample;

[0014] Adjust the parameters of the gain model based on the loss until a trained gain model is obtained.

[0015] In a third aspect, an embodiment of the present application provides a recommendation method, including:

[0016] Obtaining characteristic information corresponding to a third user at multiple preset moments, the characteristic information including identification information and user characteristics of the third user, the identification information being used to at least indicate a moment for recommending a preset interaction event with the third user as the corresponding preset moment;

[0017] The gain model is used to predict the third probability and fourth probability of the third user at each preset moment based on the feature information corresponding to each preset moment, wherein the third probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at the preset moment; the fourth probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at a random moment other than the preset moment; the gain model is trained based on the model training method of the present application;

[0018] Obtaining a probability gain at the same preset moment based on a difference between the third probability and the fourth probability at the same preset moment;

[0019] Based on the probability gain of each preset moment, a moment for recommending a preset interaction event with the third user is determined.

[0020] In a fourth aspect, an embodiment of the present application provides a data processing device, including:

[0021] a first determining unit configured to determine characteristic data of a first user and a second user, wherein the characteristic data includes user characteristics of the user to which the characteristic data belongs, a first time at which a preset interaction event is performed with the user, and an interaction result corresponding to the preset interaction event; the first time corresponding to the first user is a time at which the preset interaction event is recommended for the first user; and the first time corresponding to the second user is different from the time at which the preset interaction event is recommended for the second user;

[0022] a second determining unit, configured to determine first characteristic data and second characteristic data corresponding to a preset moment from the characteristic data, wherein the first moment in the first characteristic data is equal to the preset moment, and the first moment in the second characteristic data is different from the preset moment;

[0023] The identification unit is used to identify the first characteristic data and the second characteristic data respectively, and obtain a first sample belonging to the experimental group and a second sample belonging to the control group at a preset moment as sample data of the gain model.

[0024] In a fifth aspect, an embodiment of the present application provides a model training device, comprising:

[0025] A sample acquisition unit, configured to acquire sample data of a gain model to be trained, wherein the sample data includes a first sample belonging to an experimental group and a second sample belonging to a control group at at least one preset time, determined by the data processing method of the present application;

[0026] a prediction unit, configured to predict, based on the first sample and the second sample, respectively, a first probability of a user in the first sample and a second probability of a user in the second sample using a gain model, wherein the first probability is a probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a preset time, and the second probability is a probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a random time other than the preset time;

[0027] a loss calculation unit, configured to calculate the loss of the gain model based on the interaction result and the first probability in the first sample, and the interaction result and the second probability in the second sample;

[0028] The training unit is used to adjust the parameters of the gain model based on the loss until a trained gain model is obtained.

[0029] In a sixth aspect, an embodiment of the present application provides a recommendation device, including:

[0030] an acquiring unit, configured to acquire characteristic information corresponding to a third user at a plurality of preset moments, the characteristic information including identification information and user characteristics of the third user, the identification information being used to at least indicate a moment at which a preset interaction event with the third user is recommended to be the corresponding preset moment;

[0031] A prediction unit, configured to predict, using a gain model and based on feature information corresponding to each preset moment, a third probability and a fourth probability of a third user at each preset moment, wherein the third probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at the preset moment; and the fourth probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at a random moment other than the preset moment; the gain model is trained based on the model training method of the present application;

[0032] a gain determining unit, configured to obtain a probability gain at the same preset moment based on a difference between the third probability and the fourth probability at the same preset moment;

[0033] The recommendation unit is configured to determine a recommended time for the third user to perform a preset interaction event based on the probability gain of each preset time.

[0034] In the seventh aspect, an embodiment of the present application also provides a computer device, including a memory storing multiple instructions; a processor loads instructions from the memory to execute the steps of any data processing method, model training method, or recommendation method provided in the embodiment of the present application.

[0035] In an eighth aspect, an embodiment of the present application also provides a computer-readable storage medium, which stores a plurality of instructions, and the instructions are suitable for a processor to load to execute the steps of any data processing method, model training method, or recommendation method provided in the embodiment of the present application.

[0036] In the ninth aspect, an embodiment of the present application also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the steps of any data processing method, model training method, or recommendation method provided in the embodiment of the present application.

[0037] By adopting the scheme of the embodiment of the present application, after determining the feature data of the first user and the second user, the first feature data and the second feature data corresponding to the preset moment can be determined from the feature data according to the first moment in the feature data, wherein the feature data includes the user characteristics of the user to which the feature data belongs, the first moment of the preset interaction event with the user, and the interaction result corresponding to the preset interaction event; and the first moment corresponding to the first user is the moment when the preset interaction event is recommended to the first user; the first moment corresponding to the second user is different from the moment when the preset interaction event is recommended to the second user; and the first moment in the first feature data is equal to the preset moment, then the first feature data can be regarded as the feature data that is recommended to the user for the preset interaction event at the preset moment and the preset interaction event is performed at the preset moment, and the first moment in the second feature data is different from the preset moment, then the second feature data can be regarded as the feature data that is recommended to the user for the preset interaction event at the preset moment. A user performs a preset interaction event, but does not perform the preset interaction event at the preset moment, but performs the characteristic data corresponding to the preset interaction event at a random moment outside the preset moment. After the first characteristic data and the second characteristic data are respectively identified, the first sample belonging to the experimental group at the preset moment and the second sample belonging to the control group can be obtained as sample data of the gain model. Therefore, the present application provides sample data of performing a preset interaction event at the preset moment and performing a preset interaction event at a random moment outside the preset moment, which is conducive to the gain model learning the gain between a certain moment and a random moment outside a certain moment. Compared with the sample data of the experimental group and the control group composed of samples of performing preset interaction events at two different moments in the related art, the sample data can better reflect the influence of timing on the response of the user group to the preset interaction event, so the sample data is more effective, which is conducive to improving the accuracy and effectiveness of the interaction timing of the preset interaction events recommended to users. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1 Schematic diagram of the application environment of the data processing method provided in the embodiment of the present application;

[0040] Figure 2 This is a schematic diagram of an application scenario of the data processing method provided in the embodiments of the present application;

[0041] Figure 3This is a schematic diagram of a principle for implementing gain model training in an embodiment of the present application;

[0042] Figure 4 This is a flow chart of an embodiment of the data processing method provided in the embodiments of the present application;

[0043] Figure 5 This is a schematic diagram of a principle for implementing the generation of the first sample and the second sample provided by an embodiment of the present application;

[0044] Figure 6 This is a schematic diagram of an embodiment flow of a method for generating the first sample and the second sample provided in the embodiments of the present application;

[0045] Figure 7 This is a schematic diagram of an embodiment flow chart of another generation scheme of the first sample and the second sample provided in the embodiments of the present application;

[0046] Figure 8 This is a flow chart of an embodiment of the model training method provided in the embodiments of the present application;

[0047] Figure 9 This is a flow chart of an embodiment of the recommended method provided in the embodiments of this application;

[0048] Figure 10 is a structural diagram of a data processing device provided in an embodiment of the present application;

[0049] Figure 11 It is a schematic diagram of the internal structure of the computer device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application. At the same time, in the description of the embodiments of the present application, the terms "first", "second", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0051] In one embodiment of the present application, the data processing method, model training method, and recommendation method can be run on a local terminal device or a server. When the data processing method, model training method, and recommendation method are run on a server, the method can be implemented and executed based on a cloud interaction system, wherein the cloud interaction system includes a server and a client device.

[0052] In order to better understand the data processing method, model training method, recommendation method, device, equipment, and product provided in the embodiments of the present application, the application environment applicable to the embodiments of the present application is described below.

[0053] See also Figure 1 , Figure 1 Schematic diagram of an application environment of the data processing method provided by an embodiment of the present application is shown. As an implementation method, the data processing method provided by the embodiment of the present application can be applied to the same electronic device. The electronic device can be Figure 1 The server 110 shown in FIG. 1 can be connected to the terminal device 120 via a network. The network is used to provide a medium for a communication link between the server 110 and the terminal device 120. The network can include various connection types, such as wired communication links, wireless communication links, etc., which are not limited in the embodiments of the present application. Alternatively, in other embodiments, the electronic device can also be a smartphone, a laptop computer, etc.

[0054] It should be understood that Figure 1 The server 110, network, and terminal device 120 are merely illustrative. Any number of servers, networks, and terminal devices may be provided as needed. For example, the server 110 may be a physical server or a server cluster consisting of multiple servers, and the terminal device 120 may be a mobile phone, tablet, desktop computer, laptop computer, or the like. It will be appreciated that embodiments of the present application may also allow multiple terminal devices 120 to access the server 110 simultaneously.

[0055] In some embodiments, the terminal device 120 may receive characteristic data of the first user and the second user. Further, the terminal device 120 may send the characteristic data of the first user and the second user to the server 110 via a network. After the server 110 receives the characteristic data of the first user and the second user, the characteristic data of the first user and the second user may be processed using the data processing method of an embodiment of the present application to obtain a first sample belonging to the experimental group and a second sample belonging to the control group at a preset time.

[0056] As another implementation, the data processing method, model training method, and recommendation method provided in the embodiments of this application can be applied to different computer devices. For example, the data processing method is applied to computer device A, the model training method is applied to computer device B, and the recommendation method is applied to computer device C. The embodiments of this application do not limit the computer devices to which the above two methods are applied.

[0057] The following is a detailed description of each step in conjunction with the accompanying drawings. In this embodiment, the execution subject is a terminal device. It should be noted that the order in which the following embodiments are described does not limit the preferred order of the embodiments. Although the flowcharts illustrate a logical order, in some cases, the steps shown or described may be performed in a different order than that shown in the accompanying drawings.

[0058] The data processing method, model training method, and recommendation method of this solution can be applied to any product that needs to determine the recommended time of a preset interaction event, such as an intelligent outbound calling system.

[0059] Among them, the intelligent outbound call system can obtain the characteristic information of the terminal user, input the characteristic information of the terminal user into the gain model, obtain the probability gain of each preset time, and determine the recommended time for the terminal user to make an outbound call based on the probability gain of each preset time.

[0060] For example, see Figure 2 , Figure 2 The application scenario of the recommendation method of this solution is shown. The application solution in this scenario includes: collecting characteristic data of the target user through the client, sending the characteristic data to the intelligent outbound calling system for processing, and obtaining a recommended time for making an outbound call to the target user. The intelligent outbound calling system then dials the target user based on the recommended time and performs other operations such as service recommendations to the target user after the call is connected.

[0061] The following is a detailed description of each step in conjunction with the accompanying drawings. In this embodiment, the execution subject is a terminal device. It should be noted that the order in which the following embodiments are described does not limit the preferred order of the embodiments. Although the flowcharts illustrate a logical order, in some cases, the steps shown or described may be performed in a different order than that shown in the accompanying drawings.

[0062] In existing technologies, the results of user interactions are often influenced by the timing of these interactions. For example, in common recommendation scenarios, to achieve deeper communication with users and promote successful recommendations, it is often necessary to consider personalized recommendation timing that suits the user to improve recommendation effectiveness and success rate. Uplift modeling is often used to evaluate the impact of certain interventions (such as specific interaction timing) on ​​user behavior.

[0063] However, in related technologies, time is typically used as a feature in sample data to train a gain model. This allows the model to estimate the recommendation success rate for users at different times, derive the success rate gain between different times, and then recommend time based on this gain. However, with this type of sample data, the model can only learn the success rate gain between different fixed moments.

[0064] See also Figure 3 In the data processing method of this embodiment, a data collection module may collect user characteristics of a first user and a second user, as well as the first moment of a preset interaction event performed by these users and the interaction results of the preset interaction event performed at the first moment, to form characteristic data for each of these users. The characteristic data includes the user characteristics of the user to which the characteristic data belongs, the first moment of the preset interaction event performed with the user, and the interaction results corresponding to the preset interaction event.

[0065] Then, the first feature data and the second feature data corresponding to the preset moment are determined from the feature data. After the first feature data and the second feature data are respectively identified, the first sample belonging to the experimental group and the second sample belonging to the control group at the preset moment can be obtained as sample data of the gain model; then, the gain model (such as S-Learner and / or T-Learner) is trained according to the first sample and the second sample to obtain a trained gain model. The trained gain model is used to output the probability gain of each preset moment corresponding to the feature data according to the user's feature data.

[0066] In this application, the first moment corresponding to the first user is the moment when the preset interaction event is recommended to the first user; and the first moment in the first feature data is equal to the preset moment, then the first feature data can be considered as feature data that uses the preset moment as the recommendation to the user for the preset interaction event and performs the preset interaction event at the preset moment, and the first moment corresponding to the second user is different from the moment when the preset interaction event is recommended to the second user, that is, the first moment corresponding to the second user is a randomly selected moment other than the moment when the preset interaction event is recommended to the second user, and the first moment in the second feature data is different from the preset moment, then the second feature data can be considered as feature data that uses the preset moment as the recommendation to the user for the preset interaction event, but the preset interaction event is not performed at the preset moment, but the preset interaction event is performed at a random moment other than the preset moment.

[0067] Please refer to Figure 4 The specific process of the data processing method may be as follows: Step 401 to Step 403, wherein:

[0068] Step 401: Determine feature data of a first user and a second user.

[0069] To facilitate understanding of this application, the relevant knowledge of Uplift modeling is first introduced here.

[0070] Uplift modeling is a modeling method used in the field of recommendation and personalized decision-making to evaluate the impact of interventions (such as coupons and advertising) on ​​user behavior. Its core is to quantify the incremental impact of interventions on user responses, that is, to evaluate the effectiveness of interventions by comparing the differences in user responses with and without the intervention. The purpose of Uplift modeling is to more accurately identify users who respond positively to specific interventions, thereby optimizing resource allocation and improving the efficiency and effectiveness of recommendations. In terms of modeling methods, Uplift models mainly come in two types: dual models and single models. Dual models model users in the experimental group and control group separately, and then take the difference in predicted results as the estimated Uplift value. Single models are suitable for binary classification scenarios and are modeled through category conversion and reconstruction of the loss function.

[0071] In this application, the first user belongs to the experimental group, and the second user belongs to the control group.

[0072] In this application, recommended times for performing preset interaction events are set for users in both the experimental and control groups. For the experimental group, the first time for performing the preset interaction event is equal to the recommended time for the first user. That is, the experimental group in this article is a user group that performs the preset interaction event according to the time recommended by the model. For the control group, the first time for performing the preset interaction event is not equal to the recommended time for the second user, and is a time randomly selected outside the recommended time. Therefore, the control group in this application can also be considered a random control group. The random control group in this article is to randomly dial users at various times, thereby ensuring that users at various times are homogeneous users.

[0073] In the alternative example, users are randomly assigned between the experimental group and the randomized control group.

[0074] According to the foregoing content, the data processing solution of this application is mainly to generate more effective data samples for the gain model, so as to train a better gain model and achieve more effective prediction of the recommendation time.

[0075] The feature data includes user features of the user to which the feature data belongs, the first moment of a preset interaction event with the user, and an interaction result corresponding to the preset interaction event.

[0076] In the embodiments of the present application, the preset interaction event may be a phone call event, an advertisement delivery event, a message push event, or other interaction events; the interaction results corresponding to the preset interaction event may include interaction success and interaction failure. This application will take the preset interaction event as an example, and the interaction results corresponding to the phone call event include successful call (including user answering the call) and failed call (including user not answering the call).

[0077] In an embodiment of the present application, user characteristics can be determined based on the user's attribute information. Optionally, the attribute information may include information on attribute dimensions such as the user's age, gender, occupation, region, interests, hobbies, and income. Feature data may also include historical interaction record information of the user's preset interaction events. The historical interaction record information may include the interaction time of the preset interaction event with the user within a historical time period and the interaction result information corresponding to the interaction time. The interaction result information may represent the characteristic information of the interaction result of the preset interaction event with the user at the specific interaction time. The characteristic information may include at least one of the following: interaction success rate, failure rate, number of successful interactions, average interaction duration, conversion result of the interaction event, and revenue information corresponding to the interaction event.

[0078] For example, for a call dialing event, the historical interaction record information may include the time of the historical call dialing for the user and the corresponding call result information, where the call result information may include whether the call was connected or not. The historical interaction record information may also include the call connection rate for the historical call dialing time in a specified period. For example, it may include the call connection rate for calls made at 9:00 a.m. during the historical call dialing time within a month, or the call connection rate for calls made between 9:00 a.m. and 10:00 a.m. during the historical call dialing time within a month.

[0079] Optionally, the moment in this application can be understood as an instantaneous time or a period. Optionally, the moment in this application can be understood as a period. For example, a period of time in a day during which a preset interactive event can be performed, such as 8:00 am to 10:00 pm, is divided into T periods, resulting in T moments, namely, moment 0, moment 1, ..., moment T-1.

[0080] In this application, the time for the preset interaction event recommended for the user can be obtained by recommendation from any model with the interaction time recommendation function, or can be set manually based on experience. This example is not limited here. Optionally, after obtaining the recommended time, the specific time for the preset interaction event for the user can be arbitrarily selected from the recommended time. For example, the time recommended to the first user is T-1, which corresponds to the time period of 9:00-10:00 in the evening. The specific time for the preset interaction event for the user can be arbitrarily selected from the time period of 9:00-10:00 in the evening. When determining the characteristic data of the user, the first moment is recorded as T-1, rather than the specific interaction time.

[0081] For example, attribute information may include the user's age and gender. The interaction result may be represented by Y, where Y=1 indicates a successful interaction result and Y=0 indicates a failed interaction result. For example, for a phone call event, if the result is connected, then Y=1, otherwise Y=0. Taking the phone call event as an example, the first moment corresponding to the phone call event is used as the moment feature call_hh, where call_hh=0, ...T-1, and T is the total number of moments. The granularity of the moment feature call_hh can be customized, for example, if each moment is 2 hours, then 9:00 am to 10:00 am is moment 0, 11:00 am to 12:00 am is moment 1, 2:00 pm to 3:00 pm is moment 2, and 4:00 pm to 5:00 pm is moment 3. This is just an example, and further examples are not described in detail. Then, based on the above, the user's characteristic data can be constructed, specifically represented by (call_hh, age, gender, ..., call_hh_1_ans_rate_1m, ..., call_hh_T-1_ans_rate_1m), where call_hh_1_ans_rate_1m represents the call connection rate at time 1 (call_hh=1) in the past month (1m), and call_hh_T-1_ans_rate_1m represents the call connection rate at time T-1 (call_hh=T-1) in the past month (1m).

[0082] The selection and setting method of the first user and the second user in the present application is not limited. For example, a random selection operation can be performed from multiple users to obtain the first user and the second user with the same number of users and no overlap.

[0083] In one embodiment, a first number of users may be obtained, a first specified number of users may be selected from the first number of users as first users, and a second specified number of users may be selected from the remaining plurality of users as second users. The first specified number and the second specified number may be the same or different. Recommended times for performing a preset interaction event may be determined for the first and second users.

[0084] For the first user, a preset interactive event such as a phone call can be performed on the first user at the recommended time, and the corresponding dialing result information can be collected to obtain feature data; for the second user, the recommended time can be not followed, but a time can be randomly selected. At the randomly selected time, a preset interactive event such as a phone call can be performed with the second user, and the corresponding dialing result information can be collected to obtain feature data.

[0085] It is understandable that for each interaction moment for the preset interaction event divided in this application, such as time 0 to T-1, a certain number of samples are required to better train the gain model. Therefore, in an optional example in this application, the recommended moment corresponding to the first user and the recommended moment corresponding to the second user cover multiple interaction moments set for the preset interaction event. Optionally, the number of first users at different interaction moments may be the same or different, and the number of second users at different interaction moments may be the same or different. At the same interaction moment, the number of first users and second users may be the same or different.

[0086] Optionally, in this example, before or after dividing the users into the experimental group and the control group, the recommended moments for performing preset interaction events can be determined for the users, and then, for each recommended moment, based on the sample quantity requirement, the first user and the second user at each recommended moment are selected.

[0087] For example, 1,000 users can be obtained, and recommended times for preset interaction events can be set for the 1,000 users. 50 users whose recommended times are between 9:00 a.m. and 10:00 a.m. are selected from the 1,000 users as first users, and 50 users whose recommended times are between 9:00 a.m. and 10:00 a.m. are selected from the remaining 875 sample users as second users.

[0088] Then, you can call the first user within the time range of 9:00 a.m. to 10:00 a.m. and collect the corresponding calling result information; you can call the second user at any time other than 9:00 a.m. to 10:00 a.m. on the same day and collect the corresponding calling result information, for example, you can call a second user at 11:00 a.m. and call another second user at 2:00 p.m.

[0089] Step 402: Determine first characteristic data and second characteristic data corresponding to a preset time from the characteristic data.

[0090] The first moment in the first characteristic data is equal to the preset moment, and the first moment in the second characteristic data is different from the preset moment.

[0091] In the present application, the preset moment can be understood as one of the multiple interaction moments divided for the preset interaction event described above, such as one of the moments 0, 1,...T-1.

[0092] It is understandable that the present application can select preset moments in sequence according to the order of interaction moments, thereby determining the first sample of the experimental group and the second sample of the control group at each preset moment.

[0093] See also Figure 5 , Figure 5 A schematic diagram of the construction of the first and second samples of the present application is provided. The first sample of the Treat group can be obtained based on the characteristic data of the randomized control group and the experimental group, and the second sample of the Control group can be obtained based on the characteristic data of the randomized control group.

[0094] That is, the first characteristic data can be determined from the characteristic data of the random control group and the experimental group, and the second characteristic data can be determined from the characteristic data of the random control group.

[0095] In this application, the method for screening the first feature data and the second feature data is not limited. In one example, the first sample and the second sample can be obtained from the feature data of users in the experimental group and the random control group, respectively. That is, the first feature data is first selected from the feature data of the first user, and the second feature data is selected from the feature data of the second user, and then the first sample and the second sample are generated.

[0096] Optionally, in a specific embodiment, the method further includes:

[0097] Determining a first user group corresponding to the preset time from the first user, wherein the time for recommending a preset interaction event to the first user in the first user group is equal to the preset time;

[0098] A second user group corresponding to the preset time is determined from the second users, wherein the time for performing the preset interaction event recommended to the second users in the second user group is equal to the preset time.

[0099] Optionally, the timing of determining the first user group and the second user group is not limited, and may be before or after generating the feature data. This example does not have any limitation on this.

[0100] In an optional example, a first user group and a second user group may be divided for each preset moment.

[0101] For example, the moment labels of the model's recommended moments for preset interaction events are obtained for all users in the random control group and the experimental group. Among them, i represents user i, Indicates the recommended time, This application can divide the customer group of the random control group into the second user group The customer base of the experimental group is divided into the first user group It's understandable that the number of Ct and Tt is T, meaning there are T first and second user groups, corresponding to the T recommendation times. Because the control and treatment groups were randomly assigned, Ct and Tt represent homogeneous populations. This satisfies the Uplift model's requirement for homogeneity in the control and treatment groups.

[0102] Correspondingly, the step of "determining the first characteristic data and the second characteristic data corresponding to the preset moment from the characteristic data" includes: determining the first characteristic data at the first moment equal to the preset moment from the characteristic data of the first user group corresponding to the preset moment; and determining the second characteristic data at the first moment not equal to the preset moment from the characteristic data of the second user group corresponding to the preset moment.

[0103] In the present application, a preset moment t may be selected from T moments each time by loop traversal to determine the first characteristic data and the second characteristic data, thereby generating a sample.

[0104] For example, by looping through the selected preset time t, the first feature data required by the experimental group (treat group) at the preset time can be determined from the feature data of the first user group at the preset time t, and the second feature data required by the control group (control group) at the preset time can be determined from the feature data of the second user group at the preset time t.

[0105] In the present application, the first feature data and the second feature data may be selected from the feature data of the second user in the random control group to generate the first sample and the second sample, thereby expanding the data volume of the training sample of the gain model.

[0106] Optionally, in one embodiment, the step of “determining the first feature data and the second feature data corresponding to the preset time from the feature data” includes:

[0107] Selecting, from the feature data of the second user, at least part of the feature data whose first moment is equal to the preset moment as the first feature data corresponding to the preset moment;

[0108] At least part of the feature data of the second user whose first moment is not equal to the preset moment is selected from the feature data of the second user as the second feature data corresponding to the preset moment.

[0109] Optionally, in the present application, for each preset moment t, the first moment in the characteristic data of the second user can be looped through, and the characteristic data whose first moment is equal to the preset moment can be taken as the first characteristic data, and the characteristic data whose first moment is not equal to the preset moment can be taken as the second data.

[0110] For example, the interaction time call_hh in the feature data of the second user is traversed, and the feature data corresponding to the interaction time call_hh!=t is determined as the second feature data of the control group, and the feature data corresponding to the dialing time call_hh=t is determined as the first feature data of the treat group.

[0111] Therefore, in this application, the characteristic data stepping of the random control group can generate training samples belonging to the control group, and can also be used to generate training samples belonging to the experimental group. That is, this application implements a scheme of constructing a corresponding homogeneous random dial control group with a random control group, expands the training samples, and can make the training of the gain model more sufficient.

[0112] Step 403 : Identify the first characteristic data and the second characteristic data respectively to obtain a first sample belonging to the experimental group and a second sample belonging to the control group at a preset time.

[0113] Optionally, the first characteristic data and the second characteristic data are identified separately in the present application, mainly to indicate the preset time corresponding to the corresponding sample and whether it is the sample of the experimental group or the sample of the control group at the preset time through the set identification information. The quantity and content of the identification information are not limited and can be composed of any type of characters.

[0114] Optionally, in one example, the step of “respectively identifying the first characteristic data and the second characteristic data to obtain a first sample belonging to the experimental group and a second sample belonging to the control group at a preset time” may include:

[0115] Setting first identification information in the first feature data corresponding to the preset time, and obtaining a first sample belonging to the experimental group at the preset time, wherein the first identification information indicates that: the time when the preset interaction event is recommended in the first feature data, and the time when the preset interaction event is actually performed are equal to the preset time;

[0116] The second identification information is set in the second characteristic data corresponding to the preset moment to obtain the second sample belonging to the control group at the preset moment; the second identification information represents: the moment when the preset interaction event is recommended in the second characteristic data is equal to the preset moment, and the moment when the preset interaction event is actually performed is a randomly selected moment outside the preset moment.

[0117] Optionally, the first identification information and the second identification information may be composed of at least one type of flag information, and this example does not limit this. For example, the first identification information and the second identification information may be composed of a type of flag information, which may simultaneously indicate that the time when the preset interaction event is recommended is equal to the identification information of the preset time, such as the preset time t, and whether the recommended time is equal to the time when the preset interaction event is actually performed. For example, if the first identification information is t-1 and the second identification information is t-0, t is the preset time (which is also the recommended time).

[0118] In one example, the feature data can be represented by setting the same flag in the first feature data and the second feature data, and performing corresponding assignments. Optionally, the step of "setting the first identification information in the first feature data corresponding to the preset time to obtain the first sample belonging to the experimental group at the preset time" may include:

[0119] A first flag is set in the first feature data corresponding to the preset time and a first flag value is assigned to the first flag, and a second flag is set and a preset time is assigned to the second flag as the first identification information, thereby obtaining a first sample belonging to the experimental group at the preset time;

[0120] Setting second identification information in the second feature data corresponding to the preset time to obtain a second sample belonging to the control group at the preset time includes:

[0121] A first flag is set in the second characteristic data corresponding to the preset time and the first flag is assigned a second flag value, and a second flag is set and the second flag is assigned a preset time as the second identification information to obtain a second sample belonging to the control group at the preset time.

[0122] Optionally, the first mark and the second mark of the present application can be understood as mark features added to the feature data, each mark feature corresponds to a feature position in the feature data, and the same marks correspond to the same feature positions.

[0123] Optionally, the first flag value and the second flag value may be set arbitrarily without limitation. In one example, the first flag value and the second flag value may be 0 or 1.

[0124] For example, a first flag feature treat_flg and a second flag feature treat_hh can be set for the first feature data, and the value of the first flag feature treat_flg is 1, and the value of the second flag feature treat_hh is t, indicating that the user needs to perform a preset interaction event at time t (i.e., the recommended time), and the preset interaction event is performed at this time, indicating that the generated sample belongs to the first sample of the treat group; a first flag feature treat_flg and a second flag feature treat_hh can be set for the second feature data, and the value of the first flag feature treat_flg is 0, and the value of the second flag feature treat_hh is t, indicating that the user needs to perform a preset interaction event at time t (i.e., the recommended time), and the preset interaction event is performed at a random time other than this time, indicating that the generated sample belongs to the second sample of the control group.

[0125] It can be understood that, through the above processing, two characteristic values ​​are added to the characteristic data to obtain the first sample and the second sample.

[0126] An optional sample example is represented as:

[0127] (treat_flg,treat_hh,call_hh,age,gender,···,call_hh_1_ans_rate_1m,···,call_hh_T_ans_rate_1m).

[0128] In one embodiment, the step of "after determining the first and second feature data corresponding to the preset moment from the feature data" may further include: deleting the first moment in the first and second feature data. Thus, the first moment at which the user actually performed the preset interaction event no longer exists in the sample data. The gain model no longer knows the actual time when the user performed the preset interaction event in the second sample of the control group. It only knows that the sample was a random sample of the preset interaction event outside the preset moment. This can improve the gain model's accuracy in predicting the gain of the preset interaction event performed at a specific moment compared to a random moment outside the specific moment.

[0129] For example, the first sample at the preset time t in this application is expressed as:

[0130] The second sample of the preset time t is expressed as:

[0131] (0,t,age,gender,···,call_hh_1_ans_rate_1m,···,call_hh_T_ans_rate_1m).

[0132] Based on the above introduction, the data processing method of this application will be further explained below by taking the preset interaction event as a phone call event as an example.

[0133] 1. Collect the data set first

[0134] Collect the call records of the second user in the randomized control group and the first user in the experimental group. Use the time corresponding to the call (the granularity can be customized, for example, one time every two hours) as the time feature cal l_hh, where cal l_hh = 0, ...T-1. T is the total number of time and the interaction result. Also collect user features such as age, gender, and the user's call completion history at each time as user features. If the interaction result is a connection, Y = 1; otherwise, Y = 0.

[0135] The user's characteristic data is expressed as (cal l_hh, age, gender,…,cal l_hh_1_ans_rate_1m,…,cal l_hh_T_ans_rate_1m).

[0136] For the characteristics of the randomized control group, the Figure 6 The process shown in the figure extracts the first sample and the second sample respectively.

[0137] 2. Based on the characteristic data of the random control group, the first sample and the second sample are obtained.

[0138] like Figure 5 As shown, first, the preset time t can be determined. Then, the dialing time cal l_hh in the feature data of the random control group is looped through, and the identification feature treat_flg used to indicate whether the sample belongs to the control group or the treat group is added (treat_flg is 0 for control samples and 1 for treat samples), and the intervention identifier treat_hh corresponding to the dialing time is added.

[0139] Set the sample corresponding to the dialing time cal l_hh = t as the treat group (assign treat_flg a value of 1), and set treat_hh = t (indicates that the dialing time intervention corresponds to the treat group). Delete the dialing time feature cal l_hh from the feature data, and keep other features unchanged to obtain the first sample.

[0140] Set the sample corresponding to the dialing time cal l_hh! = t as the control group (assign treat_flg a value of 0), and set treat_hh = t (indicating that the dialing time t corresponds to the control group). Delete the dialing time feature cal l_hh from the feature data, leaving other features unchanged, to obtain the second sample.

[0141] 3. Generate the first sample and the second sample at the preset time from the characteristic data of the experimental group and the random control group respectively.

[0142] Divide the users of the control group into the first user group Divide the users of the experimental group into the second user group

[0143] See also Figure 7 , looping through different preset times t, adding the identification feature treat_flg to the feature data of the first user group corresponding to the preset time t and assigning it a value of 1, and the identification feature treat_hh, and assigning it a value of t. treat_flg = 1, indicating that the first user group at the preset time Tt (whose actual call time cal l_hh must be t, because the actual call time is consistent with the recommended time) is the treat group, and treat_hh = t, indicating that the call time t is the corresponding treat group. Finally, the call time feature cal l_hh is deleted from the feature data, while other features remain unchanged, to obtain the first sample.

[0144] All samples in the second user group Ct with the actual dialing time cal l_hh! = t are marked as treat_flg = 0, and treat_hh = t (which indicates the control group corresponding to the intervention dialing time t). The dialing time feature cal l_hh is deleted from the feature data, while other features remain unchanged, to obtain the second sample.

[0145] In this application, samples are represented as: (treat_flg, treat_hh, age, gender, ..., call_hh_1_ans_rate_1m, ..., call_hh_T_ans_rate_1m). treat_flg ranges from 0 to 1, where 0 indicates the sample belongs to the control group and is the second sample, and 1 indicates the sample belongs to the treat group and is the second sample. treat_hh ranges from 0 to T-1, corresponding to the treat / control samples intervened at time points 0 to T-1.

[0146] In the above detailed example, the order of steps 2 and 3 is not limited. In actual application, at least one of the solutions in steps 2 and 3 can be selected to generate the first sample and the second sample.

[0147] In the present application, after obtaining the first sample and the second sample, the gain model can be trained to obtain a trained gain model.

[0148] In summary, the embodiment of the present application can determine the first feature data and the second feature data corresponding to the preset moment from the feature data after determining the feature data of the first user and the second user according to the first moment in the feature data, wherein the feature data includes the user characteristics of the user to which the feature data belongs, the first moment of the preset interaction event with the user, and the interaction result corresponding to the preset interaction event; and the first moment corresponding to the first user is the moment of recommending the preset interaction event to the first user; the first moment corresponding to the second user is different from the moment of recommending the preset interaction event to the second user; and the first moment in the first feature data is equal to the preset moment, then the first feature data can be regarded as the feature data of recommending the preset interaction event to the user at the preset moment and the preset interaction event is performed at the preset moment, and the first moment in the second feature data is different from the preset moment, then the second feature data can be regarded as the feature data of recommending the preset interaction event to the user at the preset moment. A preset interaction event is performed, but the preset interaction event is not performed at the preset moment, but the characteristic data corresponding to the preset interaction event is performed at a random moment outside the preset moment. After the first characteristic data and the second characteristic data are respectively identified, the first sample belonging to the experimental group at the preset moment and the second sample belonging to the control group can be obtained as sample data of the gain model. Therefore, the present application provides sample data of performing a preset interaction event at the preset moment and performing a preset interaction event at a random moment outside the preset moment, which is conducive to the gain model learning the gain between a certain moment and a random moment outside a certain moment. Compared with the sample data of the experimental group and the control group composed of samples of preset interaction events at two different moments in the related art, the sample data can better reflect the influence of timing on the response of the user group to the preset interaction event, so the sample data is more effective, which is conducive to improving the accuracy and effectiveness of the interaction timing of the preset interaction events recommended to users.

[0149] This application leverages the random assignment properties of the dataset to construct homogeneous treatment and control groups. Furthermore, the control group represents the concept of random dialing at other times corresponding to the intervention at time t, consistent with real-world scenarios. This allows the model to output the gain generated by random dialing at each time t compared to other times, maximizing practical application satisfaction.

[0150] By fully understanding the experimental and randomized control groups, this application found a solution to construct a treat group based on the experimental group and a corresponding homogeneous random dial control group based on the randomized control group. This expanded the training sample and made the training more comprehensive. Furthermore, it was possible to more fully learn the specific gain effects of dialing at the time determined by the model, rather than being limited to random dialing.

[0151] Based on the above introduction, the following examples will further illustrate the model training method of this application. For example, please refer to Figure 8 The specific process of the model training method can be as follows: Step 801 to Step 804, wherein:

[0152] Step 801: Obtain sample data of the gain model to be trained, wherein the sample data includes a first sample belonging to the experimental group and a second sample belonging to the control group at at least one preset moment determined by the data processing method of the present application.

[0153] In the embodiment of the present application, the type of the uplift model is not limited, for example, it can be an S-Learner and / or T-Learner model.

[0154] Among them, S-Learner is a causal inference method that can be used to deal with causal effect estimation problems. The basic idea of ​​S-Learner is to estimate causal effects by establishing a single gain model. S-Learner combines the intervention variable (treat_flg in this application) with other features to train a single machine learning model (M). The model predicts the results (Y) with and without the intervention factor (control group treat_flg = 0, experimental group treat_flg = 1). The difference in results is then estimated to obtain the gain corresponding to the intervention.

[0155] T-Learner is also a causal inference method that can be used to deal with causal inference problems. T-Learner is a boosting modeling technique that treats the experimental group and the control group as separate experiments. It trains two separate sub-models—one for the experimental group and the other for the control group—enabling it to capture how the intervention affects each group differently.

[0156] Step 802: Based on the first sample and the second sample, the gain model is used to predict the first probability of users in the first sample and the second probability of users in the second sample, respectively, wherein the first probability is the probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at the preset time, and the second probability is the probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a random time other than the preset time.

[0157] Specifically, the first sample and the second sample can be input into the gain model, and the sub-model in the gain model can use the treat_flg in the feature data of the first sample and the second sample as the intervention field to output the probability of the call being connected for the first sample and the second sample under treat_hh=t in the call dialing event.

[0158] In the embodiment of the present application, the S-Learner gain model is used as an example to illustrate that the first sample of the treat group at the preset time j is input into the gain model to obtain the first probability (yt)^_ij corresponding to the treat group, and the second sample of the control group at the preset time j is input into the gain model to obtain the second probability (yc)^_ij corresponding to the control group. Then the gain corresponding to the intervention is

[0159] Taking T-Learner as an example, the first sample of the treat group at the preset time j is input into the sub-model corresponding to the treat group to obtain the first probability (yt)^_ij corresponding to the treat group. The second sample of the control group at the preset time j is input into the sub-model corresponding to the control group to obtain the second probability (yc)^_ij corresponding to the control group. The gain corresponding to the intervention is

[0160] Step 803: Calculate the loss of the gain model according to the interaction result and the first probability in the first sample, and the interaction result and the second probability in the second sample.

[0161] Optionally, the loss function used to calculate the loss in this example is not limited, for example, it can be the mean squared error loss function (MSE), the mean absolute error loss function (MAE), the Huber loss function, the Softmax loss function, etc.

[0162] Step 804: Adjust the parameters of the gain model based on the loss until a trained gain model is obtained.

[0163] After adjusting the model parameters based on the loss, the application can continue to iterate the model training until the training completion conditions are met, such as the number of training times reaches a preset threshold, or the loss converges to below the preset loss value.

[0164] After the gain model of the present application is trained, the model can be tested based on the inference data to determine whether the function of the model meets the requirements.

[0165] The following is an example of how to construct inference data.

[0166] For user i who needs to make a call later, user characteristics are collected. For the S-Learner model, inference samples need to be generated based on the number of preset moments (i.e., recommended moments). The number of preset moments is T, and the number of inference samples is at least 2T. The inference samples also include samples from the treat and control groups. In the samples, treat_flg=0 or 1, and treat_hh is j, j=0…T-1. The 2T samples of user i are input into the S-Learner model to obtain the corresponding probabilities. The probabilities of the treat group samples and the control group samples corresponding to the preset moment j are Get gain

[0167] For T-Learner, the preset number of time moments is T, then T samples need to be generated, treat_hh is j, j = 0...T-1, each sample is input into two sub-models respectively, and the probability of the treat group sample and the control group sample corresponding to the preset time moment j output by the two sub-models can be obtained. Get gain

[0168] The gain model of the present application is trained using samples obtained by the data processing method of the present application. The gain obtained based on the gain model is not the gain of a certain moment compared to other specific moments, but the gain of a certain moment compared to other random moments, which is more in line with the actual needs of recommendation moment decision-making.

[0169] Based on the above introduction, the following examples will be given to further illustrate the prediction method of this application. For example, please refer to Figure 9 The specific process of the prediction method can be as follows: Step 901 to Step 904, wherein:

[0170] Step 901: Obtain characteristic information corresponding to a third user at multiple preset moments, where the characteristic information includes identification information and user characteristics of the third user. The identification information is at least used to indicate that the moment of recommending a preset interaction event with the third user is the corresponding preset moment.

[0171] For example, the user characteristics may include information on attribute dimensions such as the user's age, gender, occupation, region, hobbies, income, etc. The identification information may indicate that the time of the recommended phone call event with the third user is the corresponding preset time.

[0172] Step 902: Predict the third probability and fourth probability of the third user at each preset moment through the gain model based on the feature information corresponding to each preset moment, wherein the third probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at the preset moment; the fourth probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at a random moment other than the preset moment; the gain model is trained based on the model training method of the present application.

[0173] In the embodiment of the present application, the S-Learner model is used as an example. For example, the number of preset moments is T, and the identification information treat_hh in the feature information is j, j = 0...T-1. Then, treat_flg = 0 or 1 can be written to the T feature information of user i, where treat_hh is j, j = 0...T-1, respectively, to obtain 2T feature information. These feature information are input into the gain model to obtain the third probability predicted by the model for the sample with treat_flg = 1. And the fourth probability predicted for the sample with treat_flg=1

[0174] In the embodiment of the present application, the T-Learner model is taken as an example for illustration, the number of preset moments is T, the identification information treat_hh in the feature information is j, j=0...T-1, and the same feature information is input into the two sub-models of the T-Learner model respectively to obtain the third probability and the fourth probability.

[0175] Step 903: Obtain a probability gain at the same preset moment based on a difference between the third probability and the fourth probability at the same preset moment.

[0176] In the embodiment of the present application, the third probability of the samples at the same time can be based on and the fourth probability Get probability gain

[0177] Step 904: Determine a recommended time for the third user to perform the preset interaction event based on the probability gain of each preset time.

[0178] Optionally, the preset time with the largest probability gain among the preset times can be selected as the recommended time for the user, that is, the recommended time is

[0179] By using the recommendation method of this embodiment, the gain in the user response probability when performing preset interaction events with the user at each preset time compared to random times outside the preset time can be determined, thereby determining the recommended time that is most suitable for the preset interaction time with the user and improving the recommendation effect.

[0180] It should be understood that, although each step in the flowcharts involved in the above embodiments is shown in sequence as indicated by the arrows, these steps are not necessarily performed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0181] Based on the same inventive concept, the embodiments of the present application further provide a data processing device for implementing the aforementioned data processing method, and a data processing device for implementing the aforementioned data processing method. The implementation solution provided by the device is similar to the implementation solution described in the aforementioned method, so the specific limitations of the embodiments of the training device for one or more gain models and the training device for the gain model provided below can be found in the above-mentioned limitations on the data processing method, and the specific limitations are not repeated here.

[0182] This embodiment also provides a data processing device, which can be integrated into a terminal device. Figure 10 As shown, the data processing device may include:

[0183] A first determining unit 1001 is configured to determine characteristic data of a first user and a second user, wherein the characteristic data includes user characteristics of the user to which the characteristic data belongs, a first time at which a preset interaction event is performed with the user, and an interaction result corresponding to the preset interaction event; the first time corresponding to the first user is a time at which the preset interaction event is recommended for the first user; the first time corresponding to the second user is different from the time at which the preset interaction event is recommended for the second user;

[0184] A second determining unit 1002 is configured to determine first characteristic data and second characteristic data corresponding to a preset time from the characteristic data, wherein the first time in the first characteristic data is equal to the preset time, and the first time in the second characteristic data is different from the preset time;

[0185] The identification unit 1003 is used to identify the first characteristic data and the second characteristic data respectively to obtain a first sample belonging to the experimental group and a second sample belonging to the control group at a preset time.

[0186] In some embodiments, the second determination unit is used to select at least part of the feature data when the first moment is equal to the preset moment from the feature data of the second user as the first feature data corresponding to the preset moment; and to select at least part of the feature data when the first moment is not equal to the preset moment from the feature data of the second user as the second feature data corresponding to the preset moment.

[0187] In some embodiments, the second determination unit is also used to determine a first user group corresponding to a preset moment from the first user, wherein the moment for recommending a preset interaction event to the first user in the first user group is equal to the preset moment; and to determine a second user group corresponding to the preset moment from the second user, wherein the moment for recommending a preset interaction event to the second user in the second user group is equal to the preset moment.

[0188] And, the second determination unit is used to determine the first characteristic data of the first moment equal to the preset moment from the characteristic data of the first user group corresponding to the preset moment; and to determine the second characteristic data of the first moment not equal to the preset moment from the characteristic data of the second user group corresponding to the preset moment.

[0189] In some embodiments, the identification unit is configured to set first identification information in the first feature data corresponding to the preset time, to obtain a first sample belonging to the experimental group at the preset time, the first identification information indicating that: the time when the preset interaction event is recommended in the first feature data, and the time when the preset interaction event is actually performed are equal to the preset time;

[0190] The second identification information is set in the second characteristic data corresponding to the preset moment to obtain the second sample belonging to the control group at the preset moment; the second identification information represents: the moment when the preset interaction event is recommended in the second characteristic data is equal to the preset moment, and the moment when the preset interaction event is actually performed is a randomly selected moment outside the preset moment.

[0191] In some embodiments, the identification unit is used to set a first flag in the first characteristic data corresponding to the preset moment and assign the first flag to a first flag value, and set a second flag and assign the second flag to a preset moment, as the first identification information, to obtain the first sample belonging to the experimental group at the preset moment; set the first flag in the second characteristic data corresponding to the preset moment and assign the first flag to a second flag value, and set the second flag and assign the second flag to a preset moment, as the second identification information, to obtain the second sample belonging to the control group at the preset moment.

[0192] In some embodiments, the identification unit is further configured to delete the first moment in the first feature data and the second feature data after determining the first feature data and the second feature data corresponding to the preset moment from the feature data.

[0193] By using the data processing device of the present application, sample data of preset interaction events performed at preset moments and preset interaction events performed at random moments other than the preset moments can be generated, which is beneficial for the gain model to learn the gain between a certain moment and a random moment other than a certain moment. Compared with the sample data of the experimental group and the control group composed of samples of preset interaction events performed at two different moments in the related technology, the sample data can better reflect the influence of timing on the response of the user group to the preset interaction event, so the sample data is more effective, which is beneficial to improving the accuracy and effectiveness of the interaction timing of the preset interaction events recommended to users.

[0194] This embodiment further provides a model training device, which can be integrated into a terminal device or a server. The model training device may include:

[0195] A sample acquisition unit, configured to acquire sample data of a gain model to be trained, wherein the sample data includes a first sample belonging to an experimental group and a second sample belonging to a control group at at least one preset time, determined by the data processing method of the present application;

[0196] a prediction unit, configured to predict, based on the first sample and the second sample, respectively, a first probability of a user in the first sample and a second probability of a user in the second sample using a gain model, wherein the first probability is a probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a preset time, and the second probability is a probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a random time other than the preset time;

[0197] a loss calculation unit, configured to calculate the loss of the gain model based on the interaction result and the first probability in the first sample, and the interaction result and the second probability in the second sample;

[0198] The training unit is used to adjust the parameters of the gain model based on the loss until a trained gain model is obtained.

[0199] This embodiment further provides a recommendation device, which can be integrated into a terminal device or a server. The recommendation device may include:

[0200] an acquiring unit, configured to acquire characteristic information corresponding to a third user at a plurality of preset moments, the characteristic information including identification information and user characteristics of the third user, the identification information being used to at least indicate a moment at which a preset interaction event with the third user is recommended to be the corresponding preset moment;

[0201] A prediction unit, configured to predict, using a gain model and based on feature information corresponding to each preset moment, a third probability and a fourth probability of a third user at each preset moment, wherein the third probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at the preset moment; and the fourth probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at a random moment other than the preset moment; the gain model is trained based on the model training method of the present application;

[0202] a gain determining unit, configured to obtain a probability gain at the same preset moment based on a difference between the third probability and the fourth probability at the same preset moment;

[0203] The recommendation unit is configured to determine a recommended time for the third user to perform a preset interaction event based on the probability gain of each preset time.

[0204] Based on the same inventive concept, an embodiment of the present application further provides a computer device, which may be a server or a terminal device. The computer device includes a memory and a processor. The memory stores a computer program, and the processor implements the steps of the above-mentioned data processing method when executing the computer program. This implements various functions, such as:

[0205] Determining characteristic data of a first user and a second user, wherein the characteristic data includes user characteristics of the user to which the characteristic data belongs, a first time of a preset interaction event with the user, and an interaction result corresponding to the preset interaction event; the first time corresponding to the first user is a time when the preset interaction event is recommended for the first user; the first time corresponding to the second user is different from the time when the preset interaction event is recommended for the second user;

[0206] Determining first characteristic data and second characteristic data corresponding to a preset time from the characteristic data, wherein the first time in the first characteristic data is equal to the preset time, and the first time in the second characteristic data is different from the preset time;

[0207] The first characteristic data and the second characteristic data are respectively identified to obtain a first sample belonging to the experimental group and a second sample belonging to the control group at a preset time.

[0208] or,

[0209] When the processor executes the computer program, it implements the steps of the above-mentioned model training method. This enables various functions, such as:

[0210] Obtaining sample data of the gain model to be trained, wherein the sample data includes a first sample belonging to the experimental group and a second sample belonging to the control group at at least one preset time determined by the data processing method of the present application;

[0211] Based on the first sample and the second sample, the gain model is used to predict a first probability of users in the first sample and a second probability of users in the second sample, respectively, where the first probability is the probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a preset time, and the second probability is the probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a random time other than the preset time;

[0212] Calculating the loss of the gain model based on the interaction result and the first probability in the first sample, and the interaction result and the second probability in the second sample;

[0213] Adjust the parameters of the gain model based on the loss until a trained gain model is obtained.

[0214] or,

[0215] When the processor executes the computer program, it implements the steps of the above-mentioned recommended method, thereby achieving various functions, such as:

[0216] Obtaining characteristic information corresponding to a third user at multiple preset moments, the characteristic information including identification information and user characteristics of the third user, the identification information being used to at least indicate a moment for recommending a preset interaction event with the third user as the corresponding preset moment;

[0217] The gain model is used to predict the third probability and fourth probability of the third user at each preset moment based on the feature information corresponding to each preset moment, wherein the third probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at the preset moment; the fourth probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at a random moment other than the preset moment; the gain model is trained based on the model training method of the present application;

[0218] Obtaining a probability gain at the same preset moment based on a difference between the third probability and the fourth probability at the same preset moment;

[0219] Based on the probability gain of each preset moment, a recommended moment for the third user to perform the preset interaction event is determined.

[0220] The above-mentioned computer device can generate sample data of preset interaction events performed at preset moments and preset interaction events performed at random moments other than the preset moments, which is conducive to the gain model learning the gain between a certain moment and a random moment other than a certain moment. Compared with the sample data of the experimental group and the control group composed of samples of preset interaction events performed at two different moments in the related technology, the sample data can better reflect the influence of timing on the response of the user group to the preset interaction event, so the sample data is more effective, which is conducive to improving the accuracy and effectiveness of the interaction timing of the preset interaction events recommended to users.

[0221] The specific implementation of the above operations can be found in the previous embodiments and will not be repeated here.

[0222] In one embodiment, the computer device is a terminal device, for example, its internal structure diagram can be as follows Figure 11 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a data processing method is implemented. The display unit of the computer device is used to form a visually visible image, and can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse, etc.

[0223] Those skilled in the art will understand that Figure 11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0224] Based on the same inventive concept, an embodiment of the present application also provides a computer-readable storage medium, which may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0225] Since the computer program stored in the computer-readable storage medium can execute any data processing method, model training method, and recommendation method provided in the embodiments of the present application, the beneficial effects that can be achieved by any data processing method, model training method, and recommendation method provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0226] Based on the same inventive concept, embodiments of the present application further provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations of the above embodiments.

[0227] It should be noted that the object data (including but not limited to user device information, user personal information, etc.) and conversation data involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions. Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0228] Any reference to the memory, database or other media used in the various embodiments provided herein may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0229] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0230] In the above-mentioned data processing device, model training device, recommendation device, computer-readable storage medium, computer equipment, and computer program product embodiments, the description of each embodiment has its own focus. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes and beneficial effects of the above-described data processing device, model training device, recommendation device, computer-readable storage medium, computer program product, computer equipment, and their corresponding units can be referred to the description of each method in the above embodiments, and the details will not be repeated here.

[0231] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0232] The above is a detailed introduction to a data processing method, model training method, recommendation method, device, computer equipment, computer-readable storage medium and computer program product provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A data processing method, characterized in that: include: Determining feature data of a first user and a second user, wherein the feature data includes user features of the user to which the feature data belongs, a first time of a preset interaction event with the user, and an interaction result corresponding to the preset interaction event; the first time corresponding to the first user is a time when the preset interaction event is recommended for the first user; and the first time corresponding to the second user is different from the time when the preset interaction event is recommended for the second user; Determining first characteristic data and second characteristic data corresponding to a preset time from the characteristic data, wherein the first time in the first characteristic data is equal to the preset time, and the first time in the second characteristic data is different from the preset time; The first characteristic data and the second characteristic data are respectively marked to obtain a first sample belonging to the experimental group and a second sample belonging to the control group at the preset moment.

2. The data processing method according to claim 1, wherein: The determining, from the characteristic data, the first characteristic data and the second characteristic data corresponding to the preset moment, includes: Selecting, from the feature data of the second user, feature data that is equal to the preset time at the first moment as the first feature data corresponding to the preset time; From the feature data of the second user, feature data whose first moment is not equal to the preset moment is selected as second feature data corresponding to the preset moment.

3. The data processing method according to claim 1, wherein: The method further comprises: Determining a first user group corresponding to the preset time from the first users, wherein the time for recommending the preset interaction event to the first user in the first user group is equal to the preset time; Determining a second user group corresponding to the preset time from the second users, wherein the time for performing the preset interaction event recommended to the second users in the second user group is equal to the preset time; The determining, from the characteristic data, the first characteristic data and the second characteristic data corresponding to the preset moment, includes: Determining, from the feature data of the first user group corresponding to the preset moment, that the first moment is equal to the first feature data of the preset moment; From the characteristic data of the second user group corresponding to the preset moment, second characteristic data indicating that the first moment is not equal to the preset moment is determined.

4. The data processing method according to any one of claims 1 to 3, characterized in that: The step of respectively marking the first characteristic data and the second characteristic data to obtain a first sample belonging to the experimental group and a second sample belonging to the control group at the preset time includes: Setting first identification information in the first feature data corresponding to the preset time, and obtaining a first sample belonging to the experimental group at the preset time, wherein the first identification information indicates that: the time when the preset interaction event is recommended in the first feature data, and the time when the preset interaction event is actually performed are equal to the preset time; Second identification information is set in the second characteristic data corresponding to the preset moment to obtain a second sample belonging to the control group at the preset moment; the second identification information represents: the moment when the preset interaction event is recommended in the second characteristic data is equal to the preset moment, and the moment when the preset interaction event is actually performed is a randomly selected moment outside the preset moment.

5. The data processing method according to claim 4, characterized in that: The step of setting the first identification information in the first feature data corresponding to the preset time to obtain the first sample belonging to the experimental group at the preset time includes: Setting a first flag in the first feature data corresponding to the preset time and assigning the first flag a first flag value, and setting a second flag and assigning the preset time to the second flag as first identification information, to obtain a first sample belonging to the experimental group at the preset time; The step of setting the second identification information in the second feature data corresponding to the preset time to obtain the second sample belonging to the control group at the preset time includes: The first flag is set in the second characteristic data corresponding to the preset time and the first flag is assigned a second flag value, and the second flag is set and the preset time is assigned to the second flag as the second identification information, to obtain the second sample belonging to the control group at the preset time.

6. The data processing method according to any one of claims 1 to 3, characterized in that: After determining the first characteristic data and the second characteristic data corresponding to the preset time from the characteristic data, the method further includes: The first moment in the first feature data and the second feature data is deleted.

7. A model training method, characterized in that: The method comprises: Obtaining sample data of a gain model to be trained, wherein the sample data includes a first sample belonging to an experimental group and a second sample belonging to a control group at at least one preset moment determined by the method according to any one of claims 1 to 6; Based on the first sample and the second sample, the gain model is used to predict a first probability of users in the first sample and a second probability of users in the second sample, respectively, wherein the first probability is the probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at the preset time, and the second probability is the probability that the user responds to the preset interaction event when the preset interaction event is performed with the user at a random time other than the preset time; Calculating the loss of the gain model according to the interaction result in the first sample and the first probability, and the interaction result in the second sample and the second probability; Parameters of the gain model are adjusted based on the loss until a trained gain model is obtained.

8. A recommendation method, characterized in that: The method comprises: Obtaining characteristic information corresponding to a third user at multiple preset moments, the characteristic information including identification information and user characteristics of the third user, the identification information being used to at least indicate a recommended moment for performing a preset interaction event with the third user as the corresponding preset moment; The gain model is used to predict the third probability and fourth probability of the third user at each preset moment based on the feature information corresponding to each preset moment, wherein the third probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at the preset moment; the fourth probability is the probability that the third user responds to the preset interaction event when the preset interaction event is performed with the third user at a random moment other than the preset moment; the gain model is trained based on the model training method according to claim 7; Obtaining a probability gain for the same preset moment based on a difference between the third probability and the fourth probability for the same preset moment; Based on the probability gain of each of the preset moments, a recommended moment for the third user to perform the preset interaction event is determined.

9. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a plurality of instructions; the processor loads instructions from the memory to execute the steps of the method according to any one of claims 1 to 6, or the steps of the method according to claim 7, or the steps of the method according to claim 8.

10. A computer program product, characterized in that The computer program product comprises a computer program, and the computer program is loaded by a processor to execute the steps of the method according to any one of claims 1 to 6, or the steps of the method according to claim 7, or the steps of the method according to claim 8.

Images