Prediction method, data processing method, computing device and computer storage medium

Abstract

Embodiments of the present application provide a prediction method, a data processing method, a computing device and a computer storage medium. The prediction method comprises: obtaining first feature data of a target object; the first feature data comprises feature values of a first number of feature types; inputting the first feature data into a data processing model to obtain second feature data; the second feature data comprises feature values of a second number of feature types, and the second number is greater than the first number; the data processing model is trained by using sample values comprising the first number of feature types and sample values comprising the second number of feature types; and predicting the target object according to the second feature data to generate a prediction result. The technical solution provided by the embodiments of the present application can improve the accuracy of the prediction result.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a prediction method, a data processing method, a computing device, and a computer storage medium. Background Technology

[0002] With the development of computer and internet technologies, online transactions and other operations can be conducted via the internet. Users can engage in various online behaviors on internet platforms. Taking online transactions as an example, users can browse, share, or purchase goods or services. Therefore, in many practical applications, user behavior prediction is necessary. For instance, in online transactions, it's necessary to predict the probability of a user purchasing a particular product. Based on the prediction results, it's possible to decide whether to incentivize the user to make a purchase.

[0003] Currently, neural network models are typically used to predict user target behaviors by using feature data of various feature types as user feature data. However, in practical applications, only feature data of some feature types can usually be obtained. Related technologies often use the mode or default value as feature data of other feature types to supplement the user feature data. This data completion method results in a lack of diversity in user feature data, thus causing the accuracy of prediction results based on user feature data to still fail to meet actual needs. Summary of the Invention

[0004] This application provides a prediction method, a data processing method, a computing device, and a computer storage medium.

[0005] Firstly, this application provides a prediction method, including:

[0006] Obtain first feature data of the target object; the first feature data includes feature values ​​of a first number of feature types;

[0007] The first feature data is input into a data processing model to obtain second feature data; the second feature data includes feature values ​​of a second number of feature types, the second number being greater than the first number; the data processing model is trained using sample values ​​including the first number of feature types and sample values ​​including the second number of feature types.

[0008] Based on the second feature data, the target object is predicted, and a prediction result is generated.

[0009] Secondly, this application provides a data processing method, including:

[0010] Obtain training samples, wherein the training samples include first sample feature data consisting of feature values ​​of a first number of feature types of the target object and second sample feature data consisting of feature values ​​of a second number of feature types.

[0011] The data processing model is trained by using the first sample data as the model input data and the second sample data as the model label data.

[0012] The data processing model is used to predict the behavior of the target object.

[0013] Thirdly, this application provides a prediction method, including:

[0014] Obtain first feature data related to the user and the target product. The first feature data includes data related to the user's purchase intention of the target product. The first feature data includes feature values ​​of a first number of feature types.

[0015] The initial feature data is input into the data processing model, and the second feature data is output. The second feature data includes feature values ​​of a second number of feature types, and the second number is greater than the first number. The data processing model is trained using sample values ​​including a first number of feature types and sample values ​​including a second number of feature types.

[0016] Using the second feature data, the probability of the user purchasing the target product is predicted.

[0017] Fourthly, embodiments of this application provide a prediction device, comprising:

[0018] The data acquisition module is used to acquire first feature data of the target object; the first feature data includes feature values ​​of a first number of feature types;

[0019] A data input module is used to input the first feature data into a data processing model to obtain second feature data; the second feature data includes feature values ​​of a second number of feature types, the second number being greater than the first number; the data processing model is trained using sample values ​​including a first number of feature types and sample values ​​including a second number of feature types.

[0020] The result generation module is used to predict the target object based on the second feature data and generate a prediction result.

[0021] Fifthly, this application provides a data processing apparatus, comprising:

[0022] The sample acquisition module is used to acquire training samples, which include first sample feature data consisting of feature values ​​of a first number of feature types of the target object and second sample feature data consisting of feature values ​​of a second number of feature types.

[0023] The training module is used to train the data processing model by taking the first sample data as the model input data and the second sample data as the model label data.

[0024] The data processing model is used to predict the behavior of the target object.

[0025] Sixthly, embodiments of this application provide a prediction device, comprising:

[0026] The feature acquisition module is used to acquire first feature data related to the user and the target product. The first feature data includes data related to the user's purchase intention of the target product, and the first feature data includes feature values ​​of a first number of feature types.

[0027] The data completion module is used to input the initial feature data into the data processing model and output the second feature data, which includes feature values ​​of a second number of feature types, the second number being greater than the first number; the data processing model is trained using sample values ​​including a first number of feature types and sample values ​​including a second number of feature types.

[0028] The prediction module is used to predict the probability of the user purchasing the target product using the second feature data.

[0029] In a seventh aspect, embodiments of this application provide a computing device, including a processing component and a storage component;

[0030] The storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the prediction method provided in the embodiments of the present invention, or to implement the data processing method provided in the embodiments of the present invention.

[0031] Eighthly, this application provides a computer storage medium storing a computer program. When the computer executes the program, it implements the prediction method provided in this embodiment of the invention, or the data processing method provided in this embodiment of the invention.

[0032] This application embodiment employs a technical solution that involves obtaining first feature data of a target object; the first feature data includes feature values ​​of a first number of feature types; inputting the first feature data into a data processing model to obtain second feature data; the second feature data includes feature values ​​of a second number of feature types, where the second number is greater than the first number; the data processing model is trained using sample values ​​of the first number of feature types and sample values ​​of the second number of feature types; and predicting the target object based on the second feature data to generate a prediction result. Before predicting the behavior of the target object, the first feature data of the target object is first supplemented using a pre-trained data processing model. Since the data processing model is trained using sample values ​​of the first number of feature types and sample values ​​of the second number of feature types, the second feature data predicted by the data processing model is closer to the true feature values ​​of the target object. Therefore, when predicting behavior based on the second feature data, a more accurate prediction result can be generated.

[0033] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart illustrating the prediction method provided in an embodiment of the present invention is shown schematically.

[0036] Figure 2 This illustration shows a schematic diagram of predicting a target object based on second feature data and generating a prediction result, provided by an embodiment of the present invention.

[0037] Figure 3 This schematic diagram illustrates a first feature data input data processing model provided in an embodiment of the present invention to obtain second feature data.

[0038] Figure 4 The diagram illustrates a prediction method provided in an embodiment of the present invention.

[0039] Figure 5 A flowchart illustrating the data processing method provided in an embodiment of the present invention is shown schematically.

[0040] Figure 6 The diagram illustrates a training data processing model provided in an embodiment of the present invention.

[0041] Figure 7 A flowchart illustrating a prediction method provided in another embodiment of the present invention is shown schematically;

[0042] Figure 8 A block diagram of the prediction device provided in an embodiment of the present invention is shown schematically;

[0043] Figure 9 A block diagram of a data processing apparatus provided in an embodiment of the present invention is shown schematically.

[0044] Figure 10 A block diagram of a prediction device provided in another embodiment of the present invention is shown schematically;

[0045] Figure 11 A block diagram of a computing device provided in an embodiment of the present invention is shown schematically. Detailed Implementation

[0046] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0047] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0048] For businesses, efficiently identifying target customers is a crucial objective for business development. Typically, businesses can use behavioral prediction of potential customers to determine the probability of a potential customer purchasing the product. Based on this probability, they can then determine whether to implement incentive measures to encourage the potential customer to make a purchase.

[0049] Currently, neural network models are typically used to predict user target behaviors by using feature data of various feature types as user feature data. However, in practical applications, only feature data of some feature types can usually be obtained. Related technologies often use the mode or default value as feature data of other feature types to supplement the user feature data. This data completion method results in a lack of diversity in user feature data, thus causing the accuracy of prediction results based on user feature data to still fail to meet actual needs.

[0050] To at least partially address the technical problems existing in related technologies, embodiments of the present invention provide a prediction method, including obtaining first feature data of a target object; the first feature data includes feature values ​​of a first number of feature types; inputting the first feature data into a data processing model to obtain second feature data; the second feature data includes feature values ​​of a second number of feature types, the second number being greater than the first number; the data processing model is trained using sample values ​​including the first number of feature types and sample values ​​including the second number of feature types; and predicting the target object based on the second feature data to generate a prediction result. Using the prediction method provided by the embodiments of the present invention, before predicting the behavior of the target object, the first feature data of the target object is first completed using a pre-trained data processing model. Since the data processing model is trained using sample values ​​including the first number of feature types and sample values ​​including the second number of feature types, the second feature data predicted by the data processing model is closer to the true feature values ​​of the target object, thus generating a more accurate prediction result when predicting behavior based on the second feature data.

[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] Figure 1 A flowchart illustrating the prediction method provided in an embodiment of the present invention is shown, such as... Figure 1 As shown, the prediction method may include the following steps:

[0053] 101. Obtain the first feature data of the target object; the first feature data includes feature values ​​of a first number of feature types.

[0054] 102. Input the first feature data into the data processing model to obtain the second feature data; the second feature data includes feature values ​​of a second number of feature types, the second number being greater than the first number; the data processing model is trained using sample values ​​including the first number of feature types and sample values ​​including the second number of feature types.

[0055] 103. Based on the second feature data, predict the target object and generate the prediction result.

[0056] According to embodiments of the present invention, the first feature data may include feature data related to the target object and the execution of the target behavior. Specifically, the first feature data may include feature data related to the target object and the counterpart of the target behavior being executed, but is not limited thereto; the first feature data may also include feature data characterizing the features of the target object itself.

[0057] According to embodiments of the present invention, the target behavior may include internet-initiated online interactive behaviors, such as reading or purchasing behaviors initiated through a predetermined platform. Furthermore, the counterparty to which the target behavior is performed may include an article to be read corresponding to the reading behavior, or a product to be purchased corresponding to the purchasing behavior.

[0058] According to embodiments of the present invention, the data processing model may include a model constructed based on deep learning algorithms and / or machine learning algorithms. After the data processing model to be trained is constructed, it can be trained using sample values ​​including a first number of feature types and sample values ​​including a second number of feature types. Training may include iteratively adjusting the model parameters of the data processing model to be trained until a preset training stopping condition is met, thereby generating a trained data processing model.

[0059] According to an embodiment of the present invention, for a first sample value including a first number of feature types, the first number of feature types cannot fully reflect the features of the target object, that is, the first sample value has at least some feature types missing.

[0060] For a second sample value that includes a second number of feature types, the second number of feature types can fully reflect the characteristics of the target object; that is, the second sample value is the complete feature data of the target object. The second sample value can include the first sample value and the third sample value. The third sample value can include the feature types missing from the first sample value. By using the third sample value to complete the first sample value, a second sample value that fully reflects the characteristics of the target object can be generated.

[0061] According to an embodiment of the present invention, when training a data processing model to be trained, a first sample value can be input into the data processing model to be trained. The data processing model to be trained predicts and generates a fourth sample value based on the first sample value. Then, the second sample value is used as the training label information, and the network parameters of the data processing model to be trained are iteratively adjusted with minimizing the difference between the second sample value and the fourth sample value as the training constraint, until the difference between the fourth sample value and the second sample value meets the preset condition, and the trained data processing model is generated.

[0062] According to an embodiment of the present invention, since the data processing model is trained and generated using the first sample value and the second sample value, after the data processing model obtains the first feature data as the input value, it can predict the missing feature type of the first feature data and generate second feature data including a second number of feature types. Furthermore, the predicted and generated second feature type data can more realistically reflect the characteristics of the target object.

[0063] According to an embodiment of the present invention, after the data processing model outputs the second feature data, prediction can be made based on the second feature data to generate a prediction result. Since the second feature type data generated by the prediction can reflect the characteristics of the target object more realistically, the accuracy of the prediction result generated based on the second feature data is high.

[0064] According to an embodiment of the present invention, since the target object typically generates continuous network interactions, the characteristic data of the target object changes as these interactions occur, which in turn leads to a corresponding change in the probability of performing the target behavior on the target object. Therefore, in response to the dynamically changing probability of target behavior execution, another embodiment of the present invention provides a data processing method, specifically:

[0065] At predetermined time intervals, the prediction results corresponding to multiple second feature data generated by the target object are statistically analyzed.

[0066] By using multiple second feature data corresponding to the prediction results, the second feature prediction data of the target object and its corresponding prediction results are predicted within a predetermined time range.

[0067] According to an embodiment of the present invention, in order to improve the efficiency of behavior prediction, the prediction results corresponding to multiple second feature data generated by the target object can be statistically analyzed at daily intervals. However, it is not limited to this. The time interval can be flexibly selected by those skilled in the art according to the actual application requirements. For example, it can be at the hourly interval, the weekly interval, the monthly interval, etc.

[0068] According to an embodiment of the present invention, the predetermined time range may include a predetermined time range after the current time, such as the next week, the next month, the next quarter, etc.

[0069] According to an embodiment of the present invention, the prediction results corresponding to multiple second feature data generated by the target object at predetermined time intervals can be specifically implemented as follows:

[0070] According to the predetermined time interval, the first time period before the predetermined time is divided into multiple time steps.

[0071] The prediction results corresponding to the second feature data generated by the target object at multiple time steps are statistically analyzed.

[0072] The second feature data and their prediction results corresponding to multiple time steps are combined in chronological order to form the first time series data.

[0073] According to embodiments of the present invention, the time step can be a more granular time division standard than a predetermined time interval, that is, a time interval can be divided into multiple time steps. For example, when the time interval is hours, the time step can be minutes; when the time interval is weeks, the time step can be days, hours, minutes; when the time interval is months, the time step can be weeks, days, hours, minutes, etc.

[0074] In a preferred embodiment of the present invention, the time interval can be a month, and the time step can be a day. Thus, with a time step of days, a month preceding a predetermined time can be divided into thirty days, and the second feature data and prediction results corresponding to each of the thirty days can be statistically analyzed.

[0075] According to an embodiment of the present invention, the first time series can be generated, for example, in the following form: L1 = {(T1:Y1); (T2:Y2)...(T...Y1)} n Y n L1 can represent the first time series, T can represent the second feature data, Y can represent the prediction result, and n can represent the number of time steps.

[0076] Using multiple prediction results, predicting the target object's outcome within a predetermined time frame can be specifically achieved as follows:

[0077] Using the first time series data, predict the second feature prediction data and the corresponding prediction result of the target object in the second time period after the predetermined time.

[0078] According to an embodiment of the present invention, the first time series data consists of the second feature data and prediction results corresponding to each time step in the historical period. Thus, based on the second feature data of the target object in the historical period, the second feature prediction data of the target object at each corresponding time step in the future period and the corresponding probability of performing the target behavior can be predicted.

[0079] According to an embodiment of the present invention, in order to further improve the prediction results of the execution target behavior corresponding to each corresponding time step in the future period of the target object, the prediction of the second feature prediction data and its corresponding prediction results of the target object in the second time period after the predetermined time using the first time series data can be specifically implemented as follows:

[0080] Set the first time span;

[0081] Perform the following iteratively at least once:

[0082] Obtain candidate data that matches the first time span from the first time series;

[0083] Using candidate data, predict the second feature prediction data corresponding to the first time point in the second time period and the corresponding prediction result;

[0084] Determine whether the iteration stopping condition is met based on the first time point;

[0085] In response to the iteration stopping condition not being met, the first time point is determined as the predetermined time.

[0086] In response to the iteration stopping condition being met, a second time series is generated.

[0087] According to an embodiment of the present invention, the above method can be used to predict the second feature prediction data and its corresponding prediction results at the first time point using the second feature data and its corresponding prediction results at K time steps prior to the first time point.

[0088] According to an embodiment of the present invention, the first time span may include K time steps, such as 30 days, 15 days, and 10 days.

[0089] According to an embodiment of the present invention, the second feature data and its prediction result corresponding to each of the K time steps can be obtained from the first time series, and the second feature prediction data and its corresponding prediction result at the first time point can be predicted accordingly.

[0090] According to an embodiment of the present invention, after generating the second feature prediction data and its corresponding prediction result at the first time point, it can be determined whether the iteration stop condition is met based on whether the first time point is the last time point in the second time period. If not, the first time point can be determined as a predetermined time, so that the prediction of the second time point after the first time point can continue until the time point of the current iteration round is the last time point in the second time period, the iteration ends, and the prediction result corresponding to each time point in the second time period is obtained to generate the second time series.

[0091] The following will elaborate on the specific implementation process of using first time series data to predict the second feature prediction data of a target object in a second time period after a predetermined time and the corresponding prediction result. It should be noted that the following examples are only used to illustrate an optional implementation of the embodiments of the present invention, and are not intended to impose any improper limitations on the embodiments of the present invention.

[0092] After obtaining the first time series, a first time span can be set, which can be less than or equal to a predetermined time interval. For example, if the predetermined time interval is 30 days, the first time span can be 30 days, 20 days, 10 days, etc. For ease of explanation, the following explanation assumes that the first time span is equal to the predetermined time interval, both being 30 days.

[0093] Since the first time span is equal to the predetermined time interval, obtaining candidate data matching the first time span from the first time series can specifically involve obtaining all data in the first time series, i.e., the second feature data and its prediction results corresponding to each day of the 30 days. Then, the obtained second feature data and its prediction results corresponding to each day of the 30 days can be used to predict the second predicted feature data and its prediction results for the first time point in the second time period. The first time point can be the first time step in the second time period. For example, if the second time period is 15 days, then the first time point can be the first day t1 of the 15 days.

[0094] Then, based on whether t1 is the last day of the 15 days, we can determine whether the iteration stopping condition is met. Obviously, it is not met at this time. Therefore, t1 can be determined as the predetermined time, and the data that matches the first time span before the predetermined time can be obtained. That is, the data corresponding to the last 29 days and t1 in the first time series can be obtained. Based on this, the second prediction feature data of the second day t2 in the 15 days and its prediction result can be predicted until the 15th day t in the 15 days is obtained. 15 The second predictive feature data and its prediction results, and t1 to t 15 The corresponding prediction results form the second time series.

[0095] According to an embodiment of the present invention, predicting the target object based on the second feature data and generating the prediction result can be specifically implemented as follows:

[0096] The second feature data is input into a pre-trained prediction model, which outputs the prediction result. The prediction model includes multiple learners.

[0097] The second feature data is input into the pre-trained prediction model, and the output prediction results include...

[0098] The second feature data is input into multiple learners, and the first prediction result corresponding to each learner is output.

[0099] The prediction results are generated based on multiple first prediction results.

[0100] Figure 2 The illustration shows a schematic diagram of predicting a target object based on second feature data and generating a prediction result, provided by an embodiment of the present invention.

[0101] like Figure 2 As shown, 201 can represent a prediction model, where the prediction model can include multiple learners 2011.

[0102] After generating the second feature data, the second feature data can be input into multiple learners 2011, and each learner 2011 can output its own calculated first prediction result. Furthermore, the multiple first prediction results can be fused using a fusion layer 2012 to generate a final prediction result.

[0103] In one embodiment of the present invention, the fusion layer may be constructed based on a fully connected layer, but is not limited thereto. The fusion layer may also calculate the average value of multiple first prediction results or calculate a weighted average value to determine the prediction result.

[0104] According to embodiments of the present invention, in related technologies, when predicting the behavior of a target object, a single model is usually used to calculate the feature data of the target object to determine the final prediction result. However, a single model often cannot adapt to a wide variety of feature data, resulting in insufficient accuracy of the prediction result.

[0105] In this embodiment of the invention, by integrating the outputs of multiple learners to determine the final prediction result through ensemble learning, the accuracy of the prediction result can be improved. Furthermore, the robustness and generalization ability of the prediction model can also be improved.

[0106] According to an embodiment of the present invention, inputting the first feature data into a data processing model to obtain the second feature data can be specifically implemented as follows:

[0107] The first feature data is parsed to determine the first data portion within the first feature data.

[0108] Generate a mask matrix based on the first data portion.

[0109] The first feature data and the mask matrix are input into the data processing model, and the second feature data is output.

[0110] According to an embodiment of the present invention, the first data portion may include a feature type missing portion of the first feature data compared to the second feature data.

[0111] According to an embodiment of the present invention, when generating the first feature data, the feature value of the feature type of the target object actually obtained can be recorded as the real value obtained. For feature types that are not obtained, their values ​​can be recorded as 0 or empty values, and the recorded real values ​​and 0 values ​​are combined to form the first feature data.

[0112] According to an embodiment of the present invention, after parsing the first feature data, the portion of the first feature data that is recorded as a value of 0 can be determined as the first data portion, which is the portion that needs to be predicted and completed, and the portion of the first feature data that is recorded as a true value can be determined as the second data portion.

[0113] Figure 3 The illustration shows a schematic diagram of the first feature data input data processing model provided in an embodiment of the present invention, which obtains the second feature data.

[0114] exist Figure 3 In the above, 301 can represent the first feature data, which includes the true value and the value 0. First, a mask matrix 302 can be generated based on the data dimension of the first feature data. Then, in the mask matrix 302, the data bits that record the value 0 in the first feature data are set to 0, and the data bits that record the true value in the first feature data are set to 1.

[0115] According to an embodiment of the present invention, the first feature data further includes a second data portion.

[0116] The process of inputting the first feature data and the mask matrix into the data processing model and outputting the second feature data can be specifically implemented as follows:

[0117] The data processing model predicts the matrix values ​​of the mask matrix based on the second data part, generating a predicted mask matrix.

[0118] The first data portion is filled with data using a prediction mask matrix to generate the second feature data.

[0119] According to an embodiment of the present invention, the generation of the second feature data can be represented by the following formula (1):

[0120]

[0121] in, This represents the second feature data. Let M represent the mask matrix, Z represent the first data part, and G(·) represent the data processing model.

[0122] According to an embodiment of the present invention, since the matrix value prediction of the prediction mask matrix is ​​based on the second data portion of the first feature data that records the true value, the matrix value of the generated prediction mask matrix is ​​closer to the true feature value of the target object. Compared with the method of completing by mode or default value in related technologies, the accuracy of the prediction result generated based on the second feature data is improved.

[0123] According to an embodiment of the present invention, the prediction method further includes:

[0124] Determine the contribution of at least one feature value in the second feature data to the prediction result.

[0125] A prompt message is generated based on the contribution level.

[0126] Output a prompt message.

[0127] According to an embodiment of the present invention, an attribution algorithm (shap value) can be used to calculate the prediction result to determine the contribution of at least one feature value in the second feature data to the prediction result.

[0128] According to an embodiment of the present invention, after determining the contribution level, a prompt message can be generated and output to the relevant information recipients of the target object.

[0129] According to an embodiment of the present invention, the recipient of the prompt information can be a salesperson who aims to sell goods to a target audience. After receiving the prompt information, the salesperson can intuitively understand the probability that the target audience will make a purchase, as well as the basis for predicting the probability, thereby increasing the salesperson's confidence in the prediction results.

[0130] Figure 4 The diagram illustrates a prediction method provided by an embodiment of the present invention.

[0131] like Figure 4 As shown, after obtaining the first feature data, it can be input into the data processing model 401 to complete the first feature data and output the second feature data. After generating the second feature data, it can be input into multiple learners of the prediction model 402. Each learner is input with its own predicted first value, and the multiple first predicted values ​​are input into the fully connected layer to output the prediction result.

[0132] After determining the prediction results, the prediction results corresponding to multiple second feature data generated by the target object can be statistically analyzed at predetermined time intervals to generate a first time series. Then, the first time series is input into the time series model 403 to predict the prediction results corresponding to each time step in the second time period after the current time, thereby generating a second time series.

[0133] On the other hand, the prediction results can be input into the attribution model 404 constructed using the attribution algorithm to generate interpretable labels, which record the contribution of at least one feature value in the second feature data to the prediction results.

[0134] Then, recommendation prompts can be generated based on interpretability tags and the second time series. These prompts are then input into the sales personnel 405 associated with the target audience, enabling them to follow up with the target audience based on the prompts and thus incentivize them to make a purchase.

[0135] Figure 5 A flowchart illustrating the data processing method provided in an embodiment of the present invention is shown, such as... Figure 5 As shown, the data processing method may include the following steps:

[0136] 501. Obtain training samples. The training samples include first sample feature data consisting of feature values ​​of a first number of feature types of the target object and second sample feature data consisting of feature values ​​of a second number of feature types.

[0137] 502. The data processing model is trained by using the first sample data as the model input data and the second sample data as the model label data.

[0138] Among them, the data processing model is used to predict the behavior of the target object.

[0139] According to embodiments of the present invention, the data processing model may include a model constructed based on deep learning algorithms and / or machine learning algorithms. After the data processing model to be trained is constructed, it can be trained using sample values ​​including a first number of feature types and sample values ​​including a second number of feature types. Training may include iteratively adjusting the model parameters of the data processing model to be trained until a preset training stopping condition is met, thereby generating a trained data processing model.

[0140] According to an embodiment of the present invention, the training data processing model can be specifically implemented by using the first sample data as model input data and the second sample data as model label data.

[0141] The first sample data is input into the data processing model to be trained, and the predicted feature data is output.

[0142] The data processing model to be trained is trained by using the constraint of minimizing the difference between the predicted feature data and the second sample feature data, thus generating the trained data processing model.

[0143] According to an embodiment of the present invention, for a first sample feature data including a first number of feature types, the first number of feature types cannot fully reflect the features of the target object, that is, the first sample feature data has at least some feature types missing.

[0144] For the second sample feature data, which includes a second number of feature types, the second number of feature types can fully reflect the characteristics of the target object; that is, the second sample feature data is the complete feature data of the target object. The second sample feature data can include the first sample feature data and the third sample feature data. The third sample feature data can include the feature types missing from the first sample feature data. By using the third sample feature data to complete the first sample values, the second sample feature data that fully reflects the characteristics of the target object can be generated.

[0145] According to an embodiment of the present invention, when training a data processing model to be trained, first sample feature data can be input into the data processing model to be trained. The data processing model to be trained predicts and generates fourth sample feature data based on the first sample feature data. Then, the second sample feature data is used as training label information, and the network parameters of the data processing model to be trained are iteratively adjusted with the goal of minimizing the difference between the second sample feature data and the fourth sample feature data as the training constraint, until the difference between the fourth sample feature data and the second sample feature data meets the preset condition, and the trained data processing model is generated.

[0146] Figure 6 The diagram illustrates a training data processing model provided in an embodiment of the present invention.

[0147] like Figure 6 As shown, 601 can represent the first sample feature data, 602 can represent the second sample feature data, and the data processing model can include a generator 603 and a discriminator 604.

[0148] According to an embodiment of the present invention, the first sample feature data may include a first data portion, which may include a feature type missing portion of the first sample feature data compared to the second sample feature data.

[0149] According to an embodiment of the present invention, when generating the first sample feature data, the feature values ​​of the feature types of the target object actually obtained can be recorded as the real values ​​obtained. For feature types that are not obtained, their values ​​can be recorded as 0 or empty values, and the recorded real values ​​and 0 values ​​are combined to form the first sample feature data.

[0150] According to an embodiment of the present invention, the first sample feature data can be parsed, and the portion of the first feature data that records a value of 0 can be determined as the first data portion, which is the portion that needs to be predicted and completed. The portion of the first feature data that records the true value is determined as the second data portion. Then, a mask matrix 605 can be generated according to the data dimension of the first sample feature data. In the mask matrix 605, the data bits that record a value of 0 in the first feature data are set to 0, and the data bits that record the true value in the first feature data are set to 1.

[0151] In the first stage, the first sample data 601 is masked by a matrix 605 and input into the generator 603. The generator 603 makes predictions based on the first sample data 601, generating predicted feature data 606. After generating the predicted feature data 606, the loss value between the predicted feature data and the second sample data 602 can be calculated, and the network parameters of the generator 603 are iteratively adjusted according to the loss value until the loss value is less than a first preset threshold. At this point, the generator 603 is trained.

[0152] After the generator 603 has been trained, in the second stage, the network parameters of the generator 603 can be fixed to prevent further changes in order to train the discriminator 604. Specifically, the first sample data 601 is input into the trained generator 603, which outputs predicted feature data 606. Then, the predicted feature data 606 and the pre-generated random variable matrix 607 are input into the discriminator 604, which outputs a probability matrix 608. Each probability value in the probability matrix 608 can characterize whether the discriminator 604 determines whether the value corresponding to that value in the predicted feature data 605 was predicted by the generator 603.

[0153] After generating probability matrix 608, the following loss function formula (2) can be optimized using probability matrix 607 until the trained data processing model is obtained.

[0154]

[0155] Where G represents the generator and D represents the discriminator. Let M represent the predicted feature data, H represent the mask matrix, and H represent the random variable matrix.

[0156] According to an embodiment of the present invention, before predicting the behavior of a target object, the prediction feature data generated by the data processing model can be cleaned first to remove data that is not important for behavior prediction.

[0157] According to an embodiment of the present invention, the training samples further include first label information;

[0158] Data processing methods also include:

[0159] Target feature data is generated based on the predicted feature data, wherein the data dimension of the target feature data is smaller than that of the predicted feature data.

[0160] According to an embodiment of the present invention, the random forest algorithm can be used to process the predicted feature data to filter out some feature data in the predicted feature data and generate target feature data.

[0161] According to embodiments of the present invention, by performing data cleaning on the predicted feature data, interfering data in the predicted feature data can be removed, thereby simplifying the subsequent calculation process and improving the model training efficiency.

[0162] Data processing methods also include:

[0163] Determine the prediction model that includes multiple learners.

[0164] The target feature data are input into the prediction model, and multiple learners are used to obtain the first prediction value for each model.

[0165] Multiple first predicted values ​​are merged to generate a second predicted value.

[0166] The prediction model is trained based on the second predicted value and the second label information.

[0167] According to embodiments of the present invention, the learner of the prediction model can be constructed and generated based on logistic regression algorithm, random forest algorithm, extreme gradient boosting algorithm, etc.

[0168] According to embodiments of the present invention, by performing ensemble learning on the prediction model, that is, by combining the first prediction values ​​of multiple learners to generate prediction results, the robustness and generalization ability of the trained prediction model can be enhanced.

[0169] According to an embodiment of the present invention, training a prediction model based on a second predicted value and second label information can be specifically implemented by training the prediction model with the constraint of minimizing the difference between the second predicted value and the second label information.

[0170] According to an embodiment of the present invention, the plurality of learners includes at least one deep learner created based on a deep learning algorithm;

[0171] Data processing methods also include:

[0172] Perturbation values ​​are added to the target feature data to generate adversarial examples.

[0173] The adversarial example is input into the deep learner, which outputs the first prediction value.

[0174] The relative entropy value is determined based on the first predicted value and the second label information.

[0175] The deep learner is trained with the constraints of maximizing the relative entropy and minimizing the difference between the first predicted value and the second label information, thus generating the trained deep learner.

[0176] According to an embodiment of the present invention, a deep learner can be trained based on the following formula (3):

[0177]

[0178] Where l represents the loss function, x represents the target feature data, y represents the second label information, α represents the weight factor, and δ represents the perturbation value.

[0179] According to an embodiment of the present invention, by adding perturbation values ​​to the target feature data and using the generated adversarial examples to perform adversarial training on the deep learner, and by using the relative entropy value as a regularization supplement to the loss function, the robustness of the trained deep learner can be increased, and the convergence and output of the deep learner can be made more stable.

[0180] According to an embodiment of the present invention, the data processing method further includes:

[0181] Identify the target model that matches the deep learner from multiple pre-trained candidate models in the model library;

[0182] Obtain the model parameters of the target model;

[0183] The model parameters are input into the deep learner so that the model parameters of the target model of the trained candidate model can be used as the initial model parameters of the deep learner.

[0184] According to embodiments of the present invention, the selection of the target model can be based on the model structure and input data structure of the deep learner. Specifically, the network structure of the selected target model can be the same as the network structure of the deep learner, and / or the format requirements of the input data of the target model can be the same as the format requirements of the input data of the deep learner.

[0185] According to an embodiment of the present invention, by transferring the model parameters of the target model, the model parameters of the already trained target model can be used as the initial model parameters of the deep learner, which enables the deep learner to converge faster, improves the training efficiency of the deep learner, and also reduces the number of samples required for training the deep learner.

[0186] According to an embodiment of the present invention, the target feature data has a time label;

[0187] Based on the difference between the second predicted value and the second label information, iteratively training multiple learners and fully connected layers can be specifically implemented as follows:

[0188] Determine the weighting factor for the difference based on the time stamp;

[0189] The target difference is determined based on the difference and weighting factors;

[0190] Multiple learners are trained iteratively based on the target difference.

[0191] According to an embodiment of the present invention, the value of the weight factor corresponding to the difference can be determined based on the time label. Specifically, the time span of the target feature data from the current time can be determined first based on the time label, and the weight factor corresponding to the target feature data with a smaller time span can be assigned a smaller value. For example, the weight factor can be assigned a value of 0-1, and a maximum time span can be set, such as 30 days. The weight factor corresponding to the maximum time span is assigned a value of 0, and the weight factor can increase proportionally as the time span decreases.

[0192] According to embodiments of the present invention, in related technologies, feature data of a target object that is a long time away from the current time is usually undersampled or directly discarded. However, feature data that is a long time away can also reflect the features of the target object to a certain extent, and undersampling or directly discarding it will result in feature loss.

[0193] According to an embodiment of the present invention, by configuring different weight factors for target feature data at different time spans from the current time, target feature data at different times can be used as training samples for the model, ensuring the differentiation of training samples and reducing feature loss.

[0194] Figure 7 A flowchart illustrating a prediction method provided in another embodiment of the present invention is shown, such as... Figure 7 As shown, the prediction method may include the following steps:

[0195] 701. Obtain first feature data related to the user and the target product. The first feature data includes data related to the user's purchase intention of the target product. The first feature data includes feature values ​​of a first number of feature types.

[0196] 702, Input the initial feature data into the data processing model and output the second feature data. The second feature data includes feature values ​​of a second number of feature types, where the second number is greater than the first number. The data processing model is trained using sample values ​​including the first number of feature types and sample values ​​including the second number of feature types.

[0197] 703. Using the second feature data, predict the probability of a user purchasing a target product.

[0198] According to an embodiment of the present invention, Figure 7 The prediction method shown is the same as Figure 1 The specific implementation methods of the prediction methods shown are the same or similar, and will not be described again here.

[0199] According to an embodiment of the present invention, the prediction method further includes:

[0200] Based on the probability of purchase, recommend target products to users, or generate prompts based on the probability of purchase and send the prompts to relevant personnel.

[0201] When businesses sell products, they typically rely on sales personnel to promote the products to users, enabling users to fully understand the products and potentially make a purchase.

[0202] In real-world applications, the number of sales personnel is often limited, creating a need to recommend potential customers to them.

[0203] In an embodiment of the present invention, by Figure 7 The prediction method shown can generate the probability of a user purchasing a target product. When the predicted purchase probability is greater than a preset probability threshold, it indicates that the user is more likely to purchase the target product. Therefore, user information and target product information can be pushed to sales personnel, enabling them to contact the user and incentivize the user to purchase the target product by issuing incentives such as red envelopes or discount coupons.

[0204] Figure 8 A block diagram of the prediction device provided in an embodiment of the present invention is illustrated schematically, such as... Figure 8 As shown, the prediction device 800 includes a data acquisition module 801, a data input module 802, and a result generation module 803.

[0205] Data acquisition module 801 is used to acquire first feature data of the target object; the first feature data includes feature values ​​of a first number of feature types;

[0206] The data input module 802 is used to input the first feature data into the data processing model to obtain the second feature data; the second feature data includes feature values ​​of a second number of feature types, the second number being greater than the first number; the data processing model is trained using sample values ​​including the first number of feature types and sample values ​​including the second number of feature types.

[0207] The result generation module 803 is used to predict the target object based on the second feature data and generate prediction results.

[0208] According to an embodiment of the present invention, the prediction device 800 further includes a statistical module and a range prediction module.

[0209] The statistics module is used to statistically analyze the prediction results corresponding to multiple second feature data generated by the target object at predetermined time intervals.

[0210] The range prediction module is used to predict the second feature prediction data and corresponding prediction results of the target object within a predetermined time range by using multiple second feature data corresponding to the prediction results.

[0211] According to an embodiment of the present invention, the statistical module includes a partitioning unit, a statistical unit, and a sequence generation unit.

[0212] The division unit is used to divide the first time period before the predetermined time into multiple time steps according to a predetermined time interval;

[0213] The statistical unit is used to statistically analyze the prediction results corresponding to the second feature data of the target object at multiple time steps.

[0214] The sequence generation unit is used to construct the first time series data by taking the second feature data corresponding to multiple time steps and their prediction results in chronological order.

[0215] The range prediction module includes a range prediction unit.

[0216] The range prediction unit is used to predict the second feature prediction data and the corresponding prediction result of the target object in the second time period after a predetermined time using the first time series data.

[0217] According to an embodiment of the present invention, the result generation module 803 includes a first input unit, a second input unit, a third input unit, and a result generation unit.

[0218] The first input unit is used to input the second feature data into the pre-trained prediction model and output the prediction result. The prediction model includes multiple learners.

[0219] The second input unit is used to input the second feature data into the pre-trained prediction model, and the output prediction results include:

[0220] The third input unit is used to input the second feature data into multiple learners respectively, and output the first prediction result corresponding to each learner respectively;

[0221] The result generation unit is used to generate a prediction result based on multiple first prediction results.

[0222] According to an embodiment of the present invention, the data input module 802 includes:

[0223] The parsing unit is used to parse the first feature data to determine the first data portion in the first feature data;

[0224] A matrix generation unit is used to generate a mask matrix based on the first data portion;

[0225] The feature generation unit is used to input the first feature data and the mask matrix into the data processing model and output the second feature data.

[0226] According to an embodiment of the present invention, the first feature data further includes a second data portion;

[0227] The feature generation unit includes a matrix value prediction unit and a filling unit.

[0228] The matrix value prediction unit is used by the data processing model to predict the matrix values ​​of the mask matrix based on the second data part, and generate a prediction mask matrix.

[0229] The padding unit is used to pad the first data portion with the prediction mask matrix to generate the second feature data.

[0230] Figure 8 The prediction device can perform Figure 1 The implementation principle and technical effects of the prediction method in the illustrated embodiments will not be repeated here. The specific methods by which each module and unit of the prediction device in the above embodiments performs its operations have been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0231] Figure 9 A block diagram of a data processing apparatus provided in an embodiment of the present invention is shown schematically, such as Figure 9 As shown, the data processing device 900 includes a sample acquisition module 901 and a training module 902.

[0232] The sample acquisition module 901 is used to acquire training samples, which include first sample feature data consisting of feature values ​​of a first number of feature types of the target object and second sample feature data consisting of feature values ​​of a second number of feature types.

[0233] Training module 902 is used to train a data processing model by using the first sample data as model input data and the second sample data as model label data.

[0234] Among them, the data processing model is used to predict the behavior of the target object.

[0235] According to an embodiment of the present invention, the training module 902 includes a first input unit and a training unit.

[0236] The first input unit is used to input the first sample data into the data processing model to be trained and output the predicted feature data.

[0237] The training unit is used to train the data processing model to be trained by minimizing the difference between the predicted feature data and the second sample feature data, and to generate the trained data-filled model.

[0238] According to an embodiment of the present invention, the training samples further include first label information;

[0239] The data processing device 900 also includes a target data generation unit.

[0240] The target data generation unit is used to generate target feature data based on the predicted feature data, wherein the data dimension of the target feature data is smaller than that of the predicted feature data.

[0241] The data processing device 900 also includes a model determination unit, a target data input unit, a fusion unit, and a prediction model training unit.

[0242] The model determination unit is used to determine the prediction model, which includes multiple learners.

[0243] The target data input unit is used to input the target feature data into the prediction model respectively, and use multiple learners to obtain the corresponding first prediction value;

[0244] The fusion unit is used to fuse multiple first predicted values ​​to generate a second predicted value;

[0245] The prediction model training unit is used to train the prediction model based on the second predicted value and the second label information.

[0246] According to an embodiment of the present invention, the plurality of learners includes at least one deep learner created based on a deep learning algorithm.

[0247] The data processing device 900 also includes an adversarial example generation unit, a first prediction unit, a relative entropy determination unit, and a deep learner training unit.

[0248] The adversarial sample generation unit is used to add perturbation values ​​to the target feature data to generate adversarial samples;

[0249] The first prediction unit is used to input adversarial examples into the deep learner and output the first prediction value.

[0250] The relative entropy determination unit is used to determine the relative entropy value based on the first predicted value and the second label information;

[0251] The deep learner training unit is used to train the deep learner with the constraints of maximizing the relative entropy and minimizing the difference between the first predicted value and the second label information, thereby generating the trained deep learner.

[0252] Figure 9 The prediction device can perform Figure 5 The implementation principle and technical effects of the data processing method in the illustrated embodiments will not be repeated here. The specific methods by which each module and unit of the data processing apparatus in the above embodiments perform operations have been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0253] Figure 10 A block diagram of a prediction device provided in another embodiment of the present invention is illustrated schematically, such as... Figure 10 As shown, the prediction device 1000 includes a sample feature acquisition module 1001, a data completion module 1002, and a prediction module 1003.

[0254] The feature acquisition module 1001 is used to acquire first feature data related to the user and the target product. The first feature data includes data related to the user's purchase intention of the target product and includes feature values ​​of a first number of feature types.

[0255] The data completion module 1002 is used to input the initial feature data into the data processing model and output the second feature data. The second feature data includes feature values ​​of a second number of feature types, and the second number is greater than the first number. The data processing model is trained using sample values ​​including the first number of feature types and sample values ​​including the second number of feature types.

[0256] The prediction module 1003 is used to predict the probability of a user purchasing a target product using the second feature data.

[0257] According to an embodiment of the present invention, the prediction device 1000 further includes a follow-up unit.

[0258] The follow-up unit is used to recommend target products to users based on the probability of purchase, or to generate prompts based on the probability of purchase and send the prompts to relevant personnel.

[0259] Figure 10The prediction device can perform Figure 6 The implementation principle and technical effects of the prediction method described in the illustrated embodiments will not be repeated here. The specific methods by which each module and unit of the prediction device in the above embodiments performs its operations have been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0260] In one possible design, the prediction device and / or data processing device provided in the embodiments of the present invention can be implemented as a computing device, such as... Figure 11 As shown, the computing device may include a storage component 1101 and a processing component 1102;

[0261] The storage component 1101 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 1102 to implement the prediction method and data processing method provided in the embodiments of the present invention.

[0262] Of course, computing devices may also include other components, such as input / output interfaces and communication components. Input / output interfaces provide an interface between processing components and peripheral interface modules, which can be output devices, input devices, etc. Communication components are configured to facilitate wired or wireless communication between the computing device and other devices.

[0263] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0264] When the computing device is a physical device, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device.

[0265] In practical applications, this computing device can be specifically deployed as a node in a message queue system, acting as a producer, consumer, relay server, or naming server in the message queue system.

[0266] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a computer, can implement the prediction method and data processing method provided in this embodiment of the invention.

[0267] This application also provides a computer program product, including a computer program that, when executed by a computer, can implement the prediction method and data processing method provided in this embodiment of the invention.

[0268] The processing component in the corresponding embodiments described above may include one or more processors to execute computer instructions to complete all or part of the steps in the method described above. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the method described above.

[0269] Storage components are configured to store various types of data to support operation within the device. Storage components can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0270] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0271] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0272] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0273] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A prediction method, characterized in that, include: Acquire first feature data of the target object; the first feature data includes feature values ​​of a first number of feature types; the first feature data includes feature data of the target object related to the execution of the target behavior, the target behavior including network interaction behavior initiated based on the Internet; The first feature data is input into a data processing model to obtain second feature data; the second feature data includes feature values ​​of a second number of feature types, the second number being greater than the first number; the data processing model is trained using sample values ​​of a first number of feature types and sample values ​​of a second number of feature types; the data processing model is a model built based on deep learning algorithms and / or machine learning algorithms, and during training, the model parameters are iteratively adjusted until a preset training stopping condition is met, with the constraint of minimizing the difference between the predicted feature data output by the model and the sample values ​​of the second number of feature types. Based on the second feature data, a prediction is made about the target object to generate a prediction result; the prediction includes a probability prediction of the target object performing the network interaction behavior.

2. The method according to claim 1, characterized in that, The method further includes: At predetermined time intervals, the prediction results corresponding to multiple second feature data generated by the target object are statistically analyzed. By using the multiple second feature data corresponding to the prediction results, the second feature prediction data of the target object and its corresponding prediction results within a predetermined time range are predicted.

3. The method according to claim 2, characterized in that, The step of statistically analyzing the prediction results corresponding to the multiple second feature data generated by the target object at predetermined time intervals includes: According to the predetermined time interval, the first time period before the predetermined time is divided into multiple time steps; The prediction results corresponding to the second feature data generated by the target object at the multiple time steps are statistically analyzed. The second feature data and their prediction results corresponding to the multiple time steps are arranged in chronological order to form the first time series data. The step of using the multiple prediction results to predict the target object within a predetermined time range includes: Using the first time series data, predict the second feature prediction data and the corresponding prediction result of the target object in the second time period after the predetermined time.

4. The method according to claim 1, characterized in that, The step of predicting the target object based on the second feature data and generating a prediction result includes: The second feature data is input into a pre-trained prediction model, and the prediction result is output. The prediction model includes multiple learners. The step of inputting the second feature data into the pre-trained prediction model and outputting the prediction result includes: The second feature data is input into the plurality of learners respectively, and a first prediction result corresponding to each learner is output. The prediction result is generated based on multiple first prediction results.

5. The method according to claim 1, characterized in that, The step of inputting the first feature data into the data processing model to obtain the second feature data includes: The first feature data is parsed to determine the first data portion within the first feature data; Generate a mask matrix based on the first data portion; The first feature data and the mask matrix are input into the data processing model, and the second feature data is output.

6. The method according to claim 5, characterized in that, The first feature data also includes a second data portion; The step of inputting the first feature data and the mask matrix into the data processing model and outputting the second feature data includes: The data processing model predicts the matrix values ​​of the mask matrix based on the second data portion to generate a predicted mask matrix. The first data portion is filled with the predicted mask matrix to generate the second feature data.

7. A data processing method, characterized in that, include: Obtain training samples, wherein the training samples include first sample feature data consisting of feature values ​​of a first number of feature types of the target object and second sample feature data consisting of feature values ​​of a second number of feature types. The data processing model is trained by using the first sample feature data as the model input data and the second sample feature data as the model label data. The data processing model is used to predict the behavior of the target object. The data processing model used to predict the behavior of the target object includes: Obtain first feature data of the target object; the first feature data includes feature values ​​of a first number of feature types; the target object includes an object that generates network interaction behavior based on the Internet, the network interaction behavior includes reading behavior and purchasing behavior based on a predetermined platform, and the first feature data includes feature data of the target object related to the execution of network interaction behavior; The first feature data is input into the data processing model to obtain the second feature data; the second feature data includes feature values ​​of a second number of feature types, the second number being greater than the first number; the data processing model is trained using sample values ​​including a first number of feature types and sample values ​​including a second number of feature types, the data processing model being a model built based on deep learning algorithms and / or machine learning algorithms, and during the training process, the model parameters are iteratively adjusted until the preset training stopping condition is met, with the constraint of minimizing the difference between the predicted feature data output by the model and the sample values ​​of the second number of feature types; Based on the second feature data, a prediction is made about the target object to generate a prediction result; the prediction includes a probability prediction of the target object performing the network interaction behavior.

8. The method according to claim 7, characterized in that, The step of using the first sample feature data as model input data and the second sample feature data as model label data to train the data processing model includes: Input the feature data of the first sample into the data processing model to be trained, and output the predicted feature data. The data processing model to be trained is trained by using the constraint of minimizing the difference between the predicted feature data and the second sample feature data, thereby generating the trained data processing model.

9. The method according to claim 8, characterized in that, The training samples also include first label information; The method further includes: Target feature data is generated based on the predicted feature data, wherein the data dimension of the target feature data is smaller than the data dimension of the predicted feature data; The method further includes: Determine a prediction model that includes multiple learners; The target feature data are respectively input into the prediction model, and the first prediction value corresponding to each of the multiple learners is obtained; The multiple first predicted values ​​are fused together to generate a second predicted value; The prediction model is trained based on the second predicted value and the second label information.

10. The method according to claim 9, characterized in that, The plurality of learners includes at least one deep learner created based on a deep learning algorithm; The method further includes: The perturbation value is added to the target feature data to generate adversarial examples; The adversarial example is input into the deep learner, and the first predicted value is output. The relative entropy value is determined based on the first predicted value and the second label information; The deep learner is trained with the constraints of maximizing the relative entropy and minimizing the difference between the first predicted value and the second label information, thereby generating the trained deep learner.

11. A prediction method, characterized in that, include: Acquire first feature data related to the user and the target product. The first feature data includes data related to the user's purchase intention for the target product. The first feature data includes feature values ​​of a first number of feature types. The data related to the user's purchase intention for the target product includes reading behavior and purchase behavior based on a pre-order platform. The first feature data is input into the data processing model, and the second feature data is output. The second feature data includes feature values ​​of a second number of feature types, where the second number is greater than the first number. The data processing model is trained using sample values ​​of a first number of feature types and sample values ​​of a second number of feature types. The data processing model is a model built based on deep learning algorithms and / or machine learning algorithms. During training, the model parameters are iteratively adjusted until a preset training stopping condition is met, with the constraint of minimizing the difference between the predicted feature data output by the model and the sample values ​​of the second number of feature types. Using the second feature data, the probability of the user purchasing the target product is predicted.

12. The method according to claim 11, characterized in that, The method further includes: Based on the purchase probability, the target product is recommended to the user, or a prompt message is generated based on the purchase probability and sent to relevant personnel.

13. A computing device, characterized in that, This includes processing components and storage components; The storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement the prediction method as described in any one of claims 1 to 6, or to implement the data processing method as described in any one of claims 7 to 10, or to implement the prediction method as described in any one of claims 11 to 12.

14. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements the prediction method as described in any one of claims 1 to 6, or the data processing method as described in any one of claims 7 to 10, or the prediction method as described in any one of claims 11 to 12.

Images