Sample splicing training method and device

Abstract

This application discloses a sample splicing training method and apparatus. The method includes: determining a first waiting time for sample splicing based on user historical behavior data, and performing sample splicing according to the first waiting time, labeling and determining positive and negative samples of the first sample data; determining a second waiting time for sample splicing, and performing sample splicing according to the second waiting time, labeling and determining positive and negative samples of the second sample data; the second waiting time is longer than the first waiting time; and using the positive and negative samples of the first and second sample data to perform mixed training on the model to obtain the trained model. By performing sample splicing using the first and second waiting times respectively to obtain the first and second sample data, and using the positive and negative samples of the first and second sample data to perform mixed training on the model, the inaccuracy of model training caused by mislabeling can be reduced.

Description

Technical Field

[0001] This application relates to the field of Internet technology, specifically to a sample splicing training method and apparatus. Background Technology

[0002] The accuracy of sample data greatly affects the accuracy of model predictions. For example, mislabeling of sample data can lead to errors in positive and negative sample identification, resulting in inaccurate model predictions.

[0003] Sample stitching involves associating and combining multiple data sets. The biggest challenge in sample data stitching lies in the trade-off between real-time performance and accuracy. Taking game sample data as an example, game installation packages are typically several gigabytes in size, and the download process is time-consuming, resulting in a significant time lag between game exposure and download completion. If sample stitching is performed after download completion, the resulting stitched sample data has 100% accuracy, but its timeliness is poor. If sample stitching is performed between exposure and download completion, mislabeling of samples may occur, leading to poor model prediction accuracy. Therefore, a sample stitching training method is needed that ensures both sample accuracy and model prediction accuracy, while also guaranteeing real-time performance. Summary of the Invention

[0004] In view of the above problems, embodiments of this application are proposed to provide a sample splicing training method and apparatus that overcomes or at least partially solves the above problems.

[0005] According to a first aspect of the embodiments of this application, a sample splicing training method is provided, comprising:

[0006] The first waiting time for sample splicing is determined based on the user's historical behavior data, and the samples are spliced ​​according to the first waiting time. The positive and negative samples of the first sample data are then labeled and determined.

[0007] Determine the second waiting time for sample assembly, and perform sample assembly based on the second waiting time, marking the positive and negative samples of the second sample data; the second waiting time is longer than the first waiting time.

[0008] The model is trained by mixing positive and negative samples from the first sample data and the second sample data to obtain the trained model.

[0009] Optionally, determining the first waiting time for sample stitching based on user historical behavior data further includes:

[0010] Obtain the first time data of the user's first historical action and the second time data of the user's second historical action, and calculate the time difference between the second time data and the first time data.

[0011] The first waiting time for sample splicing is determined based on the user's time difference.

[0012] Optionally, determining the first waiting time for sample stitching based on the user's time difference further includes:

[0013] Calculate the time difference of multiple users to obtain the time difference distribution range for each user;

[0014] The time difference in which the proportion of users in the time difference distribution interval is greater than a preset threshold is used as the first waiting time for sample splicing.

[0015] Optionally, the process of concatenating samples based on a first waiting time, and labeling and determining the positive and negative samples of the first sample data, further includes:

[0016] Based on the first waiting time, generate the first Flink sample splicing task and splice the collected sample data.

[0017] Determine whether the first sample data obtained by concatenating the samples does not contain the user's second action;

[0018] If not, label it as a negative sample of the first sample data;

[0019] If so, label it as a positive sample of the first sample data.

[0020] Optionally, determining the second waiting time for sample splicing further includes:

[0021] The time difference between the second time data and the first time data for each user is calculated, and the maximum time difference is used to determine the second waiting time for sample splicing.

[0022] Optionally, the sample splicing is performed based on the second waiting time, and the labeling of positive and negative samples for determining the second sample data further includes:

[0023] Based on the second waiting time, a second Flink sample splicing task is generated to splice the collected sample data.

[0024] Determine whether the second sample data obtained by concatenating the samples does not contain the user's second action;

[0025] If not, label it as a negative sample of the second sample data;

[0026] If so, label it as a positive sample of the second sample data.

[0027] Optionally, the model is trained by mixing positive and negative samples from the first sample data and the second sample data to obtain the trained model, which further includes:

[0028] The model is incrementally trained using positive and negative samples from the first sample data, and then fully trained using positive and negative samples from the second sample data. The training of the model using the first sample data is then calibrated to obtain the trained model.

[0029] Optionally, user historical behavior data includes user historical game operation behavior data; the first time data of the user's first historical behavior includes the time data when the user clicked on the game; the second time of the user's second historical behavior includes the time data when the user downloaded the game.

[0030] Optionally, the method further includes:

[0031] Multiple basic data are processed to obtain corresponding basic feature data;

[0032] The basic feature data is combined according to different dimensions to obtain multi-dimensional feature data; the dimensions include time dimension, label dimension and / or scene dimension.

[0033] The corresponding sample data is determined based on multi-dimensional feature data, and the sample data is divided into training set, test set or validation set; wherein the sample data is obtained by concatenating samples according to the first waiting time or the second waiting time.

[0034] According to a second aspect of the embodiments of this application, a sample splicing training apparatus is provided, comprising:

[0035] The first splicing module is adapted to determine the first waiting time for sample splicing based on the user's historical behavior data, and to perform sample splicing based on the first waiting time, and to label and determine the positive and negative samples of the first sample data.

[0036] The second stitching module is adapted to determine the second waiting time for sample stitching, and to perform sample stitching according to the second waiting time, and to label and determine the positive and negative samples of the second sample data; the second waiting time is greater than the first waiting time;

[0037] The hybrid training module is suitable for using positive and negative samples of the first sample data and positive and negative samples of the second sample data to train the model and obtain the trained model.

[0038] According to a third aspect of the embodiments of this application, a computing device is provided, including: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0039] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above sample splicing training method.

[0040] According to a fourth aspect of the present application, a computer storage medium is provided, wherein the storage medium stores at least one executable instruction, the executable instruction causing a processor to perform an operation corresponding to the sample splicing training method described above.

[0041] According to the sample splicing training method and apparatus provided in this application, sample splicing is performed using a first waiting time and a second waiting time respectively to obtain first sample data and second sample data. The positive and negative samples of the first sample data and the positive and negative samples of the second sample data are used to perform mixed training on the model, thereby reducing the inaccuracy of model training caused by mislabeling. In addition, the first waiting time determined based on the user's historical behavior data also ensures the real-time nature of the first sample data.

[0042] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0043] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0044] Figure 1 A flowchart of a sample splicing training method according to an embodiment of this application is shown;

[0045] Figure 2 A flowchart of a sample splicing training method according to another embodiment of this application is shown;

[0046] Figure 3 A schematic diagram of the structure of a sample splicing training device according to an embodiment of this application is shown;

[0047] Figure 4 A schematic diagram of the structure of a computing device according to an embodiment of this application is shown. Detailed Implementation

[0048] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0049] First, the terms and concepts involved in one or more embodiments of this application will be explained.

[0050] Sample splicing refers to the process of associating and combining various user behavior data, such as clicks, downloads, and new logins, with the feature data of candidate items during offline model training to form a complete sample.

[0051] Figure 1 A flowchart of a sample stitching training method according to an embodiment of this application is shown, as follows: Figure 1 As shown, the method includes the following steps:

[0052] Step S101: Determine the first waiting time for sample splicing based on the user's historical behavior data, and perform sample splicing based on the first waiting time, marking and determining the positive and negative samples of the first sample data.

[0053] Sample data can include data from multiple user behaviors. This data is then correlated, combined, and concatenated to obtain sample data. However, when the intervals between multiple user behaviors are long, such as 30 minutes, waiting for all behaviors to complete before concatenating the samples leads to poor real-time performance. Conversely, not waiting for the behaviors to complete can result in mislabeling. To address these issues, this embodiment determines the first waiting time for sample concatenation based on historical user behavior data, and uses this first waiting time to determine the window waiting time for sample concatenation. Historical user behavior data includes data from multiple historical user behaviors. This example uses historical game behavior data, but is not limited to this. Historical user behavior data includes the user's first and second historical behaviors. The first time data of the first historical behavior includes the time the user clicked the game, and the second time data of the second historical behavior includes the time the user downloaded the game. Sample concatenation involves combining the user's game click data and game download data to create sample data, which is then labeled as positive or negative. Based on the data on the time users clicked on the game and the time users downloaded the game, it can be statistically determined that, for example, 95% of users wait 15 minutes after clicking on the game to download and install it. Therefore, the first waiting time for sample splicing can be determined as 15 minutes. Sample splicing is performed based on the first waiting time to obtain the first sample data, and the positive and negative samples of the first sample data are labeled. For example, the data of clicking on the game and downloading it within 15 minutes is a positive sample, and the data in other cases is a negative sample.

[0054] The first sample data ensures the real-time nature of the sample data, and the first waiting time already includes the sample data of most users downloading the game. It has fewer mislabeling issues and is a low-completeness, high-time-sensitivity sample data. It can be used in real-time mode. During training, the first sample data is obtained from the message queue for training, and the trained model can be quickly applied to online services.

[0055] Step S102: Determine the second waiting time for sample splicing, and perform sample splicing according to the second waiting time, marking the positive and negative samples of the second sample data.

[0056] There may still be mislabeling in the first sample data obtained based on the first waiting time. To address this issue, this embodiment determines a second waiting time for sample splicing. The second waiting time is longer than the first waiting time. For example, the second waiting time is determined based on user historical behavior data, such as 30 minutes, to ensure the accuracy of the data.

[0057] By stitching together samples according to the second waiting time and labeling the positive and negative samples of the second sample data, it can be ensured that all the second sample data are correctly labeled, resulting in sample data with high integrity and low timeliness.

[0058] Step S103: Use the positive and negative samples of the first sample data and the positive and negative samples of the second sample data to perform mixed training on the model to obtain the trained model.

[0059] When training the model using positive and negative samples from the first sample data, some mislabeled data may exist. A small number of errors have almost no impact on model training, but the number of mislabeled first sample data accumulates over time, affecting the model's training performance. Using the second sample data for training ensures accurate model training because its positive and negative samples are completely accurate, but the training period is longer. In this embodiment, the model is trained using a mixture of positive and negative samples from both the first and second sample data. For example, the model trained on the first sample data is periodically calibrated using positive and negative samples from the second sample data, thus ensuring the model's accuracy and real-time performance.

[0060] According to the sample splicing training method provided in this application, the samples are spliced ​​using the first waiting time and the second waiting time respectively to obtain the first sample data and the second sample data. The positive and negative samples of the first sample data and the positive and negative samples of the second sample data are used to train the model in a mixed manner, which can reduce the inaccuracy of model training caused by mislabeling. In addition, the first waiting time determined based on the user's historical behavior data also ensures the real-time nature of the first sample data.

[0061] Figure 2A flowchart of a sample stitching training method according to an embodiment of this application is shown, as follows: Figure 2 As shown, the method includes the following steps:

[0062] Step S201: Obtain the first time data of the user's first historical behavior and the second time data of the user's second historical behavior, calculate the time difference between the second time data and the first time data, and determine the first waiting time for sample splicing based on the user's time difference.

[0063] For sample stitching, this embodiment first acquires the first time data of the user's first historical action and the second time data of the user's second historical action. Taking a user's game operation as an example, the first time data can be the time the user clicks the game, and the second time data can be the time the user downloads the game. The time difference between the second time data and the first time data is calculated, which is the time difference from when the user clicks the game to when the download is complete. Based on the time differences of multiple users, statistics can be performed to obtain the distribution range of time differences for each user. For example, the distribution of users corresponding to time differences of 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes, and 30 minutes can be statistically analyzed to determine the time required for a user to complete the download from clicking the game. The time difference in which the proportion of users in the time difference distribution range is greater than a preset threshold is used as the first waiting time for sample stitching. The preset threshold is selected as 95%, that is, the time difference required for the vast majority of users to complete the download from clicking the game is used as the first waiting time for sample stitching.

[0064] Step S202: Based on the first waiting time, generate the first Flink sample splicing task, splice the collected sample data, and determine whether the first sample data obtained by sample splicing does not contain the user's second behavior.

[0065] After determining the first waiting time, a first Flink sample splicing task is generated based on the first waiting time. For example, if 15 minutes is used as the Flink window waiting time, the sample data collected within 15 minutes is spliced. It is then determined whether the first sample data obtained from the sample splicing does not contain the user's second behavior, such as whether it contains the user's game download behavior. If yes, step S204 is executed; otherwise, step S203 is executed.

[0066] Step S203: Label the negative samples as the first sample data.

[0067] For the first sample data that does not contain the user's second behavior, it is labeled as a negative sample.

[0068] Step S204: Mark the positive samples as the first sample data.

[0069] The first sample data containing the user's second behavior is labeled as a positive sample.

[0070] Step S205: Calculate the time difference between the second time data and the first time data for each user, and determine the second waiting time for sample splicing by obtaining the maximum time difference.

[0071] Corresponding to the second waiting time, the time difference between the second and first time data for each user can be calculated to obtain the maximum time difference. For example, the time differences can be sorted from largest to smallest to obtain the maximum time difference, which is then used as the second waiting time for sample concatenation. The second waiting time ensures that if a user downloads the game, the download will be completed within this time, resulting in high data accuracy during sample concatenation.

[0072] Step S206: Based on the second waiting time, generate a second Flink sample splicing task, splice the collected sample data, and determine whether the second sample data obtained by sample splicing does not contain the user's second behavior.

[0073] Based on the second waiting time, a second Flink sample splicing task is generated. For example, if 30 minutes is taken as the Flink window waiting time, the sample data collected within 30 minutes is spliced ​​to obtain the second sample data.

[0074] Determine whether the second sample data does not contain the user's second behavior, such as whether it contains the user's game download behavior. If yes, proceed to step S208; otherwise, proceed to step S207.

[0075] Step S207: Label the negative samples as the second sample data.

[0076] Secondary sample data that does not contain user second behavior is labeled as a negative sample.

[0077] Step S208: Mark the positive samples as the second sample data.

[0078] For the second sample data that contains the user's second behavior, label it as a positive sample.

[0079] Step S209: Incremental training of the model is performed using positive and negative samples of the first sample data, full training of the model is performed using positive and negative samples of the second sample data, and the training of the model using the first sample data is calibrated to obtain the trained model.

[0080] After obtaining the positive and negative samples of the first sample data, incremental training of the model is performed using these samples. The first sample data can be stored in a message queue, and the data can be retrieved directly from the message queue in real-time. Real-time mode is used for incremental training of the model, resulting in a rapid training outcome. For the positive and negative samples of the second sample data, which are accurate, full training of the model can be performed. The accuracy of these samples is used to calibrate the training of the model using the first sample data, improving the model's accuracy and resulting in the trained model.

[0081] Furthermore, before inputting the sample data into the model for training, it can be divided into different datasets, such as training, testing, or validation sets. The sample data in the training set is used for model training and learning, the sample data in the validation set is used to adjust the model and evaluate its performance, and the sample data in the testing set is used to determine the model's accuracy and evaluate its final performance. Here, sample data refers to data obtained by concatenating samples based on the first or second waiting time.

[0082] The various features contained in the sample data can be obtained by processing multiple basic data sets to obtain corresponding basic feature data. Basic data includes user data, user operation data, and object data (e.g., if the object is a game, the object data includes game data). Data analysis and feature extraction are performed on the basic data to obtain basic feature data. Furthermore, the basic feature data can be combined according to different dimensions to obtain multi-dimensional feature data. Dimensions include time, label, and scene dimensions, etc., and the corresponding sample data is determined based on the multi-dimensional feature data. A unified management method can be used for all basic data, feature data, and sample data for convenient subsequent use; no specific limitations are imposed here.

[0083] According to the sample stitching training method provided in this application, based on the time difference between the first time data of a user's historical first action and the second time data of a user's historical second action, the proportion of users within the time difference distribution interval is statistically analyzed to determine the first waiting time. A first Flink sample stitching task is then generated, and positive and negative samples are labeled on the first sample data to ensure its real-time performance. The second waiting time for sample stitching is determined based on the maximum time difference between the second and first time data, generating a second Flink sample stitching task. Positive and negative samples are then labeled on the second sample data to ensure its accuracy. The model is incrementally trained using the positive and negative samples of the first sample data, and fully trained using the positive and negative samples of the second sample data. The training of the model using the first sample data is then calibrated to improve the accuracy of model training.

[0084] Figure 3 A schematic diagram of the sample splicing training device provided in one embodiment of this application is shown. Figure 3As shown, the device includes:

[0085] The first splicing module 310 is adapted to determine the first waiting time for sample splicing based on user historical behavior data, and to perform sample splicing based on the first waiting time, and to label and determine the positive and negative samples of the first sample data.

[0086] The second splicing module 320 is adapted to determine the second waiting time for sample splicing, and to perform sample splicing according to the second waiting time, and to mark and determine the positive and negative samples of the second sample data; the second waiting time is greater than the first waiting time.

[0087] The hybrid training module 330 is suitable for using positive and negative samples of the first sample data and positive and negative samples of the second sample data to perform hybrid training on the model in order to obtain the trained model.

[0088] Optionally, the first splicing module 310 is further adapted to:

[0089] Obtain the first time data of the user's first historical action and the second time data of the user's second historical action, and calculate the time difference between the second time data and the first time data.

[0090] The first waiting time for sample splicing is determined based on the user's time difference.

[0091] Optionally, the first splicing module 310 is further adapted to:

[0092] Calculate the time difference of multiple users to obtain the time difference distribution range for each user;

[0093] The time difference in which the proportion of users in the time difference distribution interval is greater than a preset threshold is used as the first waiting time for sample splicing.

[0094] Optionally, the first splicing module 310 is further adapted to:

[0095] Based on the first waiting time, generate the first Flink sample splicing task and splice the collected sample data.

[0096] Determine whether the first sample data obtained by concatenating the samples does not contain the user's second action;

[0097] If not, label it as a negative sample of the first sample data;

[0098] If so, label it as a positive sample of the first sample data.

[0099] Optionally, the second splicing module 320 is further adapted to:

[0100] The time difference between the second time data and the first time data for each user is calculated, and the maximum time difference is used to determine the second waiting time for sample splicing.

[0101] Optionally, the second splicing module 320 is further adapted to:

[0102] Based on the second waiting time, a second Flink sample splicing task is generated to splice the collected sample data.

[0103] Determine whether the second sample data obtained by concatenating the samples does not contain the user's second action;

[0104] If not, label it as a negative sample of the second sample data;

[0105] If so, label it as a positive sample of the second sample data.

[0106] Optionally, the hybrid training module 330 is further adapted to:

[0107] The model is incrementally trained using positive and negative samples from the first sample data, and then fully trained using positive and negative samples from the second sample data. The training of the model using the first sample data is then calibrated to obtain the trained model.

[0108] Optionally, user historical behavior data includes user historical game operation behavior data; the first time data of the user's first historical behavior includes the time data when the user clicked on the game; the second time of the user's second historical behavior includes the time data when the user downloaded the game.

[0109] Optionally, the device further includes: a feature management module 340, adapted to process multiple basic data to obtain corresponding basic feature data; to combine the basic feature data according to different dimensions to obtain multi-dimensional feature data; the dimensions include time dimension, label dimension and / or scene dimension; to determine the corresponding sample data according to the multi-dimensional feature data; and to divide the sample data into training set, test set or validation set; wherein the sample data is obtained by concatenating samples according to a first waiting time or a second waiting time.

[0110] The descriptions of the above modules refer to the corresponding descriptions in the method embodiments, and will not be repeated here.

[0111] According to the sample splicing training device provided in this application, the samples are spliced ​​using the first waiting time and the second waiting time respectively to obtain the first sample data and the second sample data. The positive and negative samples of the first sample data and the positive and negative samples of the second sample data are used to perform mixed training on the model, thereby reducing the inaccuracy of model training caused by mislabeling. In addition, the first waiting time determined based on the user's historical behavior data also ensures the real-time nature of the first sample data.

[0112] This application also provides a non-volatile computer storage medium storing at least one executable instruction that can execute the sample splicing training method in any of the above method embodiments.

[0113] Figure 4 The diagram shows a structural schematic of a computing device according to an embodiment of the present application. The specific embodiments of the present application do not limit the specific implementation of the computing device.

[0114] like Figure 4 As shown, the computing device may include: a processor 402, a communications interface 404, a memory 406, and a communications bus 408.

[0115] in:

[0116] The processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408.

[0117] Communication interface 404 is used to communicate with other network elements such as clients or other servers.

[0118] The processor 402 is used to execute program 410, specifically to execute the relevant steps in the above-described sample splicing training method embodiment.

[0119] Specifically, program 410 may include program code that includes computer operation instructions.

[0120] Processor 402 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement this application. The computing device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0121] Memory 406 is used to store program 410. Memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0122] Specifically, program 410 can be used to cause processor 402 to execute the sample splicing training method in any of the above method embodiments. The specific implementation of each step in program 410 can be found in the corresponding descriptions of the steps and units in the above sample splicing training embodiments, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments, and will not be repeated here.

[0123] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this application is not directed to any particular programming language. It should be understood that the content of this application described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing preferred embodiments of this application.

[0124] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0125] Similarly, it should be understood that, in order to simplify this application and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of this application, various features of this application are sometimes grouped together into a single embodiment, figure, or description thereof. However, this method of disclosure should not be construed as reflecting an intention that the claimed application requires more features than are expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of this application.

[0126] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0127] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.

[0128] The various component embodiments of this application can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components according to this application. This application can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such an implementation of this application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0129] It should be noted that the above embodiments are illustrative of this application and not restrictive, and those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.

Claims

1. A sample splicing training method, comprising: The first waiting time for sample splicing is determined based on the user's historical behavior data, and the samples are spliced ​​according to the first waiting time. The positive and negative samples of the first sample data are labeled and determined. Specifically, the first time data of the user's historical first behavior and the second time data of the user's historical second behavior are obtained, and the time difference between the second time data and the first time data is calculated. The time differences of multiple users are counted to obtain the time difference distribution interval of each user. The time difference in which the proportion of users in the time difference distribution interval is greater than a preset threshold is taken as the first waiting time for sample splicing. A second waiting time for sample splicing is determined, and sample splicing is performed according to the second waiting time. Positive and negative samples of the second sample data are marked and determined. The second waiting time is greater than the first waiting time. The time difference between the second time data and the first time data of each user is calculated, and the maximum time difference is used to determine the second waiting time for sample splicing. The model is trained by mixing positive and negative samples of the first sample data and positive and negative samples of the second sample data to obtain the trained model; wherein, the model is incrementally trained by using positive and negative samples of the first sample data, and the model is fully trained by using positive and negative samples of the second sample data, and the training of the model by the first sample data is calibrated to obtain the trained model.

2. The method according to claim 1, wherein, The step of performing sample splicing based on the first waiting time and labeling and determining the positive and negative samples of the first sample data further includes: Based on the first waiting time, a first Flink sample splicing task is generated to splice the collected sample data. Determine whether the first sample data obtained by concatenating the samples does not contain the user's second action; If not, label it as a negative sample of the first sample data; If so, label it as a positive sample of the first sample data.

3. The method according to claim 1, wherein, The step of performing sample splicing based on the second waiting time and labeling and determining the positive and negative samples of the second sample data further includes: Based on the second waiting time, a second Flink sample stitching task is generated to stitch the collected sample data together. Determine whether the second sample data obtained by concatenating the samples does not contain the user's second action; If not, label it as a negative sample of the second sample data; If so, label it as a positive sample of the second sample data.

4. The method according to any one of claims 1-3, wherein, The user's historical behavior data includes the user's historical game operation behavior data; the first time data of the user's historical first behavior includes the time data when the user clicked on the game; the second time of the user's historical second behavior includes the time data when the user downloaded the game.

5. The method according to any one of claims 1-3, wherein, The method further includes: Multiple basic data are processed to obtain corresponding basic feature data; The basic feature data is combined according to different dimensions to obtain multi-dimensional feature data; the dimensions include time dimension, label dimension and / or scene dimension. The corresponding sample data is determined based on multi-dimensional feature data, and the sample data is divided into a training set, a test set, or a validation set; wherein the sample data is obtained by concatenating samples according to a first waiting time or a second waiting time.

6. A sample splicing training device, comprising: The first splicing module is adapted to determine the first waiting time for sample splicing based on the user's historical behavior data, and to perform sample splicing based on the first waiting time, and to label and determine the positive and negative samples of the first sample data; wherein, the first time data of the user's historical first behavior and the second time data of the user's historical second behavior are obtained, and the time difference between the second time data and the first time data is calculated; the time differences of multiple users are statistically analyzed to obtain the time difference distribution interval of each user; and the time difference in which the proportion of users in the time difference distribution interval is greater than a preset threshold is taken as the first waiting time for sample splicing. The second splicing module is adapted to determine the second waiting time for sample splicing, and to perform sample splicing according to the second waiting time, and to mark and determine the positive and negative samples of the second sample data; the second waiting time is greater than the first waiting time; wherein, the time difference between the second time data and the first time data of each user is calculated, and the maximum time difference is obtained to determine the second waiting time for sample splicing; The hybrid training module is adapted to perform hybrid training on the model using positive and negative samples of the first sample data and positive and negative samples of the second sample data to obtain the trained model; wherein, the model is incrementally trained using positive and negative samples of the first sample data, the model is fully trained using positive and negative samples of the second sample data, and the training of the model using the first sample data is calibrated to obtain the trained model.

7. A computing device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the sample splicing training method as described in any one of claims 1-5.

8. A computer storage medium storing at least one executable instruction that causes a processor to perform an operation corresponding to the sample splicing training method as described in any one of claims 1-5.

9. A computer program product comprising at least one executable instruction that causes a processor to perform an operation corresponding to the sample splicing training method as described in any one of claims 1-5.

Images