Model training method and device, information prediction method and device, storage medium and equipment

Abstract

The invention relates to the technical field of computers, and relates to a model training method and device, an information prediction method and device, a computer readable storage medium and electronic equipment. The model training method comprises the steps that user sample features are input into a feature sharing network in a multi-task model, first feature representation is obtained, and the user sample features are obtained according to user features and corresponding behavior guiding features; processing the behavior guide features through an attention structure in a multi-task model, and determining a second feature representation by combining the obtained attention features and the first feature representation; and taking the second feature representation as input of a double-tower model structure in the multi-task model, performing prediction through the double-tower model structure to obtain a pre-acquisition behavior probability and a corresponding pre-acquisition amount, and performing parameter adjustment on the multi-task model based on the pre-acquisition behavior probability and the pre-acquisition amount in combination with a sample standard label. According to the invention, the prediction accuracy of the pre-acquisition behavior of the user can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and more specifically, to a model training method, an information prediction method, a model training device, an information prediction device, a computer-readable storage medium, and an electronic device. Background Art

[0002] With the development of computer technology, trained prediction models are widely used to predict user behavior in various application scenarios, and subsequent work is guided by the prediction results.

[0003] When training the prediction model in the related technology, there are long-tail distribution training samples and uneven sample distribution, which affects the prediction accuracy of the prediction model, reduces the prediction accuracy of the user's pre-acquisition behavior, and further affects the actual operation effect based on the prediction results.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The purpose of the present disclosure is to provide a model training method and device, an information prediction method and device, a computer-readable storage medium and an electronic device, thereby improving the prediction accuracy of the prediction model at least to a certain extent and improving the prediction accuracy of the user's pre-acquisition behavior.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0007] According to one aspect of the present disclosure, a model training method is provided, including: inputting user sample features into a feature sharing network in a multi-task model to obtain a first feature representation, wherein the user sample features are obtained based on user features and corresponding behavior guidance features; processing the behavior guidance features through an attention structure in the multi-task model, and determining a second feature representation in combination with the obtained attention features and the first feature representation; using the second feature representation as an input to a dual-tower model structure in the multi-task model, respectively predicting a pre-acquisition behavior probability and a corresponding pre-acquisition amount through the dual-tower model structure, and adjusting the parameters of the multi-task model based on the pre-acquisition behavior probability and the pre-acquisition amount in combination with sample standard labels.

[0008] In an exemplary embodiment of the present disclosure, the user characteristics include user attribute characteristics and user behavior characteristics, and obtaining the user sample characteristics based on the user characteristics and the corresponding behavior guidance characteristics includes: splicing the user attribute characteristics, the user behavior characteristics and the behavior guidance characteristics to obtain the user sample characteristics.

[0009] In an exemplary embodiment of the present disclosure, the dual-tower model structure includes a first prediction model and a second prediction model; the dual-tower model structure is used to respectively predict the pre-acquisition behavior probability and the corresponding pre-acquisition amount, and based on the pre-acquisition behavior probability and the pre-acquisition amount, the multi-task model is adjusted in combination with the sample standard label, including: using the first prediction model to predict the second feature representation to obtain the pre-acquisition behavior probability corresponding to the user sample feature; based on the pre-acquisition behavior probability, predicting the second feature representation according to the second prediction model to obtain the pre-acquisition amount; determining a first loss function according to the pre-acquisition behavior probability and the corresponding standard behavior label, and determining a second loss function according to the pre-acquisition amount and the corresponding standard amount label, so as to adjust the parameters in the multi-task model in combination with the first loss function and the second loss function.

[0010] In an exemplary embodiment of the present disclosure, a first loss function is determined based on the pre-acquired behavior probability and the corresponding standard behavior label, and a second loss function is determined based on the pre-acquired amount and the corresponding standard amount label, so as to adjust the parameters in the multi-task model in combination with the first loss function and the second loss function, including: if the pre-acquired amount is a non-null value, adjusting the parameters in the first prediction model and the second prediction model in combination with the first loss function and the second loss function; if the pre-acquired amount is a null value, adjusting the parameters in the first prediction model according to the first loss function.

[0011] In an exemplary embodiment of the present disclosure, before determining the second loss function based on the pre-acquired amount and the corresponding standard amount label, the method also includes: dividing the historical pre-acquired amounts of sample users into multiple groups; and determining the standard amount label corresponding to the group based on the historical pre-acquired amounts of sample users in each group.

[0012] According to one aspect of the present disclosure, there is provided an information prediction method, comprising: taking a current user feature and a first behavior guidance feature as inputs of a prediction model, predicting a first pre-acquisition behavior probability and a first pre-acquisition amount corresponding to the first pre-acquisition behavior probability; inputting the current user feature and the second behavior guidance feature into the prediction model, predicting a second pre-acquisition behavior probability and a second pre-acquisition amount corresponding to the second pre-acquisition behavior probability, the first behavior guidance feature and the second behavior guidance feature respectively corresponding to whether to perform user behavior guidance; determining a first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount, determining a second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount, and determining a predicted amount gain based on the first expected amount and the second expected amount; wherein the prediction model is obtained after training a multi-task model according to any one of the above-mentioned model training methods.

[0013] In an exemplary embodiment of the present disclosure, before determining the first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount and determining the second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount, the method further includes: calibrating the first pre-acquisition behavior probability based on a probability calibration model, and calibrating the second pre-acquisition behavior probability based on the probability calibration model.

[0014] In an exemplary embodiment of the present disclosure, determining the first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount, and determining the second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount include: in response to the first pre-acquisition behavior probability being lower than a preset probability threshold, determining that the first expected amount corresponding to the first pre-acquisition behavior probability is a null value; or, in response to the second pre-acquisition behavior probability being lower than the preset probability threshold, determining that the second expected amount corresponding to the second pre-acquisition behavior probability is a null value.

[0015] According to one aspect of the present disclosure, a model training device is provided, the device comprising: a first processing module, used to input user sample features into a feature sharing network in a multi-task model to obtain a first feature representation, wherein the user sample features are obtained based on user features and corresponding behavior guidance features; a second processing module, used to process the behavior guidance features through an attention structure in the multi-task model, and determine a second feature representation in combination with the obtained attention features and the first feature representation; a model parameter adjustment module, used to use the second feature representation as an input of a dual-tower model structure in the multi-task model, predict a pre-acquisition behavior probability and a corresponding pre-acquisition amount through the dual-tower model structure, and adjust the parameters of the multi-task model based on the pre-acquisition behavior probability and the pre-acquisition amount in combination with sample standard labels.

[0016] According to one aspect of the present disclosure, there is provided an information prediction device, comprising: a first prediction module, used to take a current user feature and a first behavior guidance feature as inputs of a prediction model, and predict a first pre-acquired behavior probability and a first pre-acquired amount corresponding to the first pre-acquired behavior probability; a second prediction module, used to input the current user feature and the second behavior guidance feature into the prediction model, and predict a second pre-acquired behavior probability and a second pre-acquired amount corresponding to the second pre-acquired behavior probability, the first behavior guidance feature and the second behavior guidance feature respectively corresponding to whether to perform user behavior guidance; a gain prediction module, used to determine a first expected amount based on the first pre-acquired behavior probability and the first pre-acquired amount, determine a second expected amount based on the second pre-acquired behavior probability and the second pre-acquired amount, and determine a predicted amount gain based on the first expected amount and the second expected amount; wherein the prediction model is obtained after training a multi-task model according to any one of the methods described above.

[0017] According to one aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, any one of the above methods is implemented.

[0018] According to one aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned methods by executing the executable instructions.

[0019] The technical solution provided in the exemplary embodiment of the present disclosure, during the model training stage, on the one hand, simultaneously obtains the pre-acquisition behavior probability and the pre-acquisition amount corresponding to the pre-acquisition behavior probability through a multi-task model, which can avoid the sample imbalance problem that exists when only predicting the pre-acquisition amount for the user; on the other hand, combines user characteristics and behavior guidance characteristics as user sample characteristics, and uses the attention structure to process the behavior guidance characteristics, and then combines the attention characteristics with the first feature representation and continues training, and realizes enhanced processing of the intermediate features of the model based on the behavior guidance characteristics, thereby increasing the performance of the model on the behavior guidance characteristics, strengthening its ability to distinguish the behavior guidance characteristics, and thus improving the user behavior prediction accuracy of the model. During the model application stage, on the one hand, the behavior guidance characteristics can be determined according to different user behavior guidance methods, and respectively combined with the current user characteristics and input into the prediction model to obtain user behavior prediction results corresponding to different user behavior guidance methods, including pre-acquisition probability behavior and corresponding pre-acquisition amount, and then the expected amount corresponding to different user behavior guidance methods is determined based on the user behavior prediction results corresponding to different user behavior guidance methods, further avoiding the sample unevenness problem that exists when directly predicting the amount, and the predicted amount gain can be obtained through the expected amount corresponding to different user behavior guidance methods. The predicted amount gain can be used to improve the targeted guidance of subsequent work and improve the accuracy of operations.

[0020] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, in which:

[0022] Figure 1 A schematic diagram showing the stages involved in the prediction scheme of the pre-acquisition behavior according to an embodiment of the present disclosure;

[0023] Figure 2 A flow chart of a model training method according to an exemplary embodiment of the present disclosure is shown;

[0024] Figure 3 A schematic diagram of the structure of a multi-task model according to an exemplary embodiment of the present disclosure is shown;

[0025] Figure 4 A schematic diagram of the structure of another multi-task model according to an exemplary embodiment of the present disclosure is shown;

[0026] Figure 5A schematic diagram of a process for adjusting a multi-task model according to an exemplary embodiment of the present disclosure is shown;

[0027] Figure 6 A schematic diagram showing a model parameter adjustment according to an exemplary embodiment of the present disclosure is shown;

[0028] Figure 7 A flowchart of an information prediction method according to an exemplary embodiment of the present disclosure is shown;

[0029] Figure 8 A flowchart of an information prediction method in a marketing application scenario according to an exemplary embodiment of the present disclosure is shown;

[0030] Fig. 9 A structural diagram of a model training device according to an exemplary embodiment of the present disclosure is shown;

[0031] Fig.10 A structural diagram of an information prediction device according to an exemplary embodiment of the present disclosure is shown;

[0032] Fig.11 A block diagram of an electronic device according to an exemplary embodiment of the present disclosure is shown.

[0033] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts. DETAILED DESCRIPTION

[0034] The exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, the exemplary embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present disclosure will be more comprehensive and complete and the concepts of the exemplary embodiments will be fully conveyed to those skilled in the art. The same reference numerals in the figures represent the same or similar structures, and thus their detailed description will be omitted.

[0035] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known structures, methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the present disclosure.

[0036] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or these functional entities or parts of functional entities may be implemented in one or more software hardened modules, or these functional entities may be implemented in different networks and / or processor devices and / or microcontroller devices.

[0037] Figure 1 A schematic diagram showing the stages involved in the prediction scheme of the user's pre-acquisition behavior according to an embodiment of the present disclosure. Figure 1 The prediction scheme of the user's pre-acquisition behavior in the embodiment of the present disclosure may include a model training phase and a model application phase.

[0038] In the model training stage, the user sample features can be determined by combining the user features and the corresponding behavior guidance features, and the multi-task model can be trained using the user sample features until the model training stop conditions, such as convergence, are met to obtain the trained prediction model. In the model application stage, the trained prediction model can be used to classify and predict the current user features and the corresponding behavior guidance features to obtain the corresponding pre-acquisition behavior probability and pre-acquisition amount, and then based on the user behavior prediction results corresponding to different user behavior guidance methods (including pre-acquisition behavior probability and pre-acquisition amount), the expected amount corresponding to different user behavior guidance methods can be determined, and the predicted amount gain can be obtained through the expected amount corresponding to different user behavior guidance methods.

[0039] Among them, in order to strengthen the influence of behavior guidance features in the prediction results, the embodiment of the present disclosure takes the current user features and the corresponding behavior guidance features as the input of the model at the same time, and in order to further avoid the sample imbalance problem that exists in only predicting the pre-acquired amount, in the model application stage, the predicted amount gain is determined according to the expected amounts corresponding to different user behavior guidance methods.

[0040] In related technologies, the user's historical behavior characteristics are modeled and trained, and the trained prediction model is used to predict user behavior, and the subsequent work is guided by the prediction results, such as product recommendations and customer marketing.

[0041] However, the prediction model has the disadvantages of long-tail distribution of training samples and uneven sample distribution during training, which affects the prediction accuracy of the prediction model. For example, in the scenario of attracting new customers and prompting them to place orders through telemarketing, the new customer group is a complete link from the activation authentication node, the login node to the order node. The customer group volume decays at each node, and there are a large number of customers with zero order amount, resulting in a long-tail distribution of the amount and unbalanced samples, which greatly affects the prediction accuracy of the trained prediction model.

[0042] Based on one or more of the above problems, in an exemplary embodiment of the present disclosure, a model training method is first provided, such as Figure 2 A flow chart of a model training method according to an exemplary embodiment of the present disclosure is shown. Figure 2 As shown, the model training method of the embodiment of the present disclosure may include steps S210 to S230:

[0043] In step 210: the user sample features are input into the feature sharing network in the multi-task model to obtain a first feature representation, where the user sample features are obtained based on the user features and the corresponding behavior guidance features.

[0044] In the exemplary embodiment of the present disclosure, user characteristics include user attribute characteristics and user behavior characteristics. User attribute characteristics refer to relevant attribute characteristics that affect the user's pre-acquisition behavior probability and pre-acquisition amount, including but not limited to gender, age, occupation, income, user-related product information (such as item information) and account information of related products (such as account information of financial products), etc. User behavior characteristics refer to relevant operation behaviors that can reflect the user's history, current or future, such as browsing behavior, transaction behavior, recommendation behavior, collection behavior, consultation behavior, etc.

[0045] For example, in the field of financial lending, user attribute characteristics may include income level and account information associated with relevant financial products. Among them, the level of user income affects their lending behavior. The high-income customer group does not have a strong demand for lending, but the amount of demand is high. The middle-income customer group has a strong demand for lending, but the amount of demand is not high. The higher the frequency of a user's browsing of financial products, the higher the user's demand for lending. The transaction behavior reflects the user's consumption level and, to a certain extent, represents the user's possibility of borrowing. The higher the consumption level, the higher the possibility of borrowing, and vice versa. Of course, there are many other features that can be used to predict the user's pre-acquisition behavior probability and pre-acquisition amount, which are not listed here one by one.

[0046] The corresponding behavior guidance feature of user characteristics refers to whether the user has received user behavior guidance. For example, the behavior guidance feature refers to the marketing feature of whether the user has been marketed or not. For another example, the behavior guidance feature refers to whether the user has obtained pre-recommended items. That is, the behavior guidance feature can be used to indicate whether there is a relevant behavior guidance investment cost for the user. For example, if the user has been marketed, it means that there is marketing investment for the user.

[0047] After obtaining user features and corresponding behavior guidance features, the user features and behavior guidance features belonging to the same sample user are spliced ​​to obtain user sample features, that is, user attribute features, user behavior features and behavior guidance features are spliced ​​to obtain user sample features as training samples input into the multi-task model, thereby avoiding the behavior guidance features from being ignored among many user features and further strengthening the behavior guidance features in the training samples.

[0048] The multi-task model includes feature sharing network, attention structure and dual tower model structure. Figure 3 FIG. 4 shows a schematic diagram of a multi-task model according to an exemplary embodiment of the present disclosure. Figure 3 , the user features and behavior-guided features are concatenated and input into the feature sharing network in the multi-task model to obtain the first feature representation.

[0049] The feature sharing network (shared bottom) in the multi-task model is used as a bottom-level shared network to realize the sharing of information between different tasks. That is to say, task information sharing is realized between different tasks of the dual-tower model structure based on the feature sharing network. The embodiment of the present disclosure does not impose any feature limitation on the feature sharing network structure.

[0050] It should be noted that before entering the feature sharing network, the user sample features can also be processed by embedding layers and other dense layers, and there is no special limitation on this.

[0051] In step S220, the behavior guidance feature is processed by the attention structure in the multi-task model, and the second feature representation is determined by combining the obtained attention feature and the first feature representation.

[0052] In an exemplary embodiment of the present disclosure, the attention structure is a network structure based on the attention mechanism, which is used to further enhance the behavior guidance features to obtain attention features.

[0053] like Figure 4 A schematic diagram of another multi-task model according to an exemplary embodiment of the present disclosure is shown below. Figure 4 The process of processing the behavior-guided features through the attention structure in the multi-task model and obtaining the second feature representation is illustrated by way of example.

[0054] First, the behavior guidance features are embedded in an embedding manner using an embedding layer, and then the embedding results are processed using a fully connected layer so that the dimension of the output intermediate features is consistent with the first feature representation output by the feature sharing network. Then, a scaling factor between 0 and 1 is obtained after sigmoid activation, namely, the attention feature. Finally, the attention feature is multiplied by the first feature representation to achieve scaling of the first feature representation and enhance the influence of the behavior guidance features.

[0055] It is worth noting that the access position of the attention structure in the multi-task model can be determined according to actual needs, and the method of combining the attention feature and the first feature representation to determine the second feature representation is not limited to multiplying the features, but can also include but is not limited to adding, splicing, etc. The embodiments of the present disclosure do not impose special restrictions on this, and the residual connection of the attention structure can also be adjusted according to actual needs, not limited to Figure 4 The structure shown.

[0056] In addition, the feature sharing network and dual-tower model structure in the multi-task model may include multiple fully connected layers, and the number of layers and neurons can be customized as needed.

[0057] In step S230, the second feature representation is used as the input of the dual-tower model structure in the multi-task model. The dual-tower model structure is used to predict the pre-acquisition behavior probability and the corresponding pre-acquisition amount. Based on the pre-acquisition behavior probability and the pre-acquisition amount, the multi-task model is adjusted in combination with the sample standard label.

[0058] In an exemplary embodiment of the present disclosure, the dual-tower model structure includes a first prediction model and a second prediction model, which are respectively used to perform different prediction tasks, and the first prediction model and the second prediction model realize the sharing of different task information. Among them, the first prediction model is a binary classification task model, which is used to determine whether to perform the pre-acquisition behavior according to the probability of the pre-acquisition behavior, and the second prediction model is a multi-classification task model, which is used to predict the corresponding pre-acquisition amount when the pre-acquisition behavior is performed.

[0059] The sample standard labels include the standard behavior labels corresponding to the binary classification task and the standard amount labels corresponding to the multi-classification task. The following is an explanation of the process of constructing the dataset and sample standard labels.

[0060] Determine the modeling users and time windows from historical user data, and divide the training set and test set. Among them, for the first prediction model, determine the binary standard behavior label corresponding to the probability of pre-acquisition behavior, including executing pre-acquisition behavior and not executing pre-acquisition behavior, which can be identified by 0 or 1. For the second prediction model, the pre-acquisition amount is divided into multiple gears and then a multi-classification prediction task is performed. Among them, the historical pre-acquisition amount of sample users can be divided into multiple groups, and the standard amount label corresponding to the group is determined based on the historical pre-acquisition amount of sample users in each group.

[0061] Specifically, the historical pre-acquisition amount can be divided into tiers according to the number of people who have pre-acquisition behavior, so that the number of people in each tier is the same, and divided into 3 tiers, 5 tiers, 7 tiers, etc. The specific number of tiers can be determined according to demand. Then, the average amount of the historical pre-acquisition amount of sample users in each tier is used as the label amount label corresponding to the tier.

[0062] After obtaining the pre-acquisition behavior probability and the corresponding pre-acquisition amount, the multi-task model can be adjusted based on the pre-acquisition behavior probability and the pre-acquisition amount in combination with the sample standard label until the model training is completed.

[0063] Figure 5 A schematic diagram of a process for adjusting a multi-task model according to an exemplary embodiment of the present disclosure is shown, Figure 5 As shown, the process may include steps S510 to S530:

[0064] Step S510: Use the first prediction model to perform prediction processing on the second feature representation to obtain the pre-acquisition behavior probability corresponding to the user sample feature.

[0065] The second feature representation is predicted based on the first prediction model to obtain the pre-acquisition behavior probability of the user sample feature, which is used to indicate the possibility of the user performing the pre-acquisition behavior, and to determine whether the user performs the pre-acquisition behavior, whether it occurs or not, based on the pre-acquisition behavior probability.

[0066] Step S520: Based on the pre-acquisition behavior probability, the second feature representation is predicted according to the second prediction model to obtain the pre-acquisition amount.

[0067] Whether to perform the pre-acquisition behavior can be determined based on the pre-acquisition behavior probability, wherein the second feature representation corresponding to the pre-acquisition behavior can be predicted using the second prediction model to obtain the corresponding pre-acquisition amount when the pre-acquisition behavior is performed. In other words, the classification task of determining the pre-acquisition amount is for the user group that has performed the pre-acquisition behavior.

[0068] Step S530: Determine a first loss function based on the pre-acquired behavior probability and the corresponding standard behavior label, and determine a second loss function based on the pre-acquired amount and the corresponding standard amount label, so as to adjust the parameters in the multi-task model in combination with the first loss function and the second loss function.

[0069] The first loss function and the second loss function can be cross entropy loss. Of course, other loss functions can also be selected according to actual needs, and there is no special limitation on this.

[0070] Since the classification task of the pre-acquisition amount is for the user group that has performed the pre-acquisition behavior, the parameter update of the second prediction model corresponding to the task only uses the data of the user group that has performed the pre-acquisition behavior. Figure 6 A schematic diagram of model parameter adjustment according to an embodiment of the present disclosure is shown, Figure 6 As shown, the process includes:

[0071] If the pre-acquired amount is a non-null value, adjusting parameters in the first prediction model and the second prediction model in combination with the first loss function and the second loss function;

[0072] If the pre-acquired amount is a null value, the parameters in the first prediction model are adjusted according to the first loss function.

[0073] The gradient descent method can be used to update the model parameters. The classification task of the pre-acquisition amount is for the user group that has performed the pre-acquisition behavior. The second prediction model corresponding to the task only uses the gradient of the user group that has performed the pre-acquisition behavior to update the model, while the user group that has not performed the pre-acquisition behavior does not participate in the parameter update. For details, see the following formulas (1) to (3):

[0074]

[0075]

[0076]

[0077] Among them, CE is the cross entropy, y1 and y2 are the outputs of the first prediction model and the second prediction model respectively. and are the sample standard labels corresponding to y1 and y2 respectively, Whether the pre-acquisition amount is zero, when the pre-acquisition amount is zero, the second prediction model is not updated, but the first prediction model is not associated with whether the pre-acquisition amount is zero.

[0078] It should be noted that the embodiment of the present disclosure adopts the basic gradient descent method to update the model parameters, and the specific update process will not be repeated. When the pre-acquired amount is a non-null value, the parameters in the first prediction model and the second prediction model are adjusted in combination with the first loss function and the second loss function, including the update of the parameters in the entire multi-task model.

[0079] Marketing, as a user behavior guidance scenario, refers to the process of discovering or exploring potential consumer needs, allowing consumers to understand items and purchase items. In the training process of the prediction model in the marketing scenario, the behavior guidance features can include marketing features of users who are marketed and users who are not marketed. The pre-acquired behavior probability is the predicted order probability, and the pre-acquired amount is the predicted order amount corresponding to the execution of the order.

[0080] By training the multi-task model, the prediction model can accurately predict the user's order probability and the corresponding order amount. Since the marketing features are strengthened during the model training process, the model's ability to distinguish marketing features is increased, thereby improving the model's prediction accuracy of user pre-acquisition behavior in marketing scenarios.

[0081] The model training method provided in the exemplary embodiment of the present disclosure, on the one hand, can avoid the sample imbalance problem that exists when only predicting the pre-acquisition amount for the user by simultaneously predicting the pre-acquisition behavior probability and the pre-acquisition amount corresponding to the pre-acquisition behavior probability through a multi-task model, and on the other hand, combines user characteristics and behavior guidance characteristics as user sample characteristics, and uses the attention structure to process the behavior guidance characteristics, and then combines the attention characteristics with the first feature representation and continues to train, and realizes enhanced processing of the intermediate features of the model based on the behavior guidance characteristics, thereby increasing the performance of the model on the behavior guidance characteristics, and strengthening its ability to distinguish the behavior guidance characteristics, thereby improving the user behavior prediction accuracy of the model.

[0082] In an exemplary embodiment of the present disclosure, a method for predicting information is also provided. Figure 7 A flowchart of an information prediction method according to an exemplary embodiment of the present disclosure is shown. Figure 7 The information prediction method of the embodiment of the present disclosure may include the following steps:

[0083] Step S710: using the current user feature and the first behavior guiding feature as inputs of the prediction model, predicting a first pre-acquisition behavior probability and a first pre-acquisition amount corresponding to the first pre-acquisition behavior probability;

[0084] Step S720: input the current user feature and the second behavior guidance feature into the prediction model, predict the second pre-acquisition behavior probability and the second pre-acquisition amount corresponding to the second pre-acquisition behavior probability, and the first behavior guidance feature and the second behavior guidance feature respectively correspond to whether to perform user behavior guidance;

[0085] Step S730: determining a first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount, determining a second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount, and determining a predicted amount gain based on the first expected amount and the second expected amount;

[0086] Among them, the prediction model is obtained by training the multi-task model according to the above model training method, which will not be repeated here.

[0087] Through the information prediction method of the exemplary embodiment of the present disclosure, on the one hand, the behavior guidance characteristics can be determined according to different user behavior guidance methods, and respectively combined with the current user characteristics and input into the prediction model to obtain user behavior prediction results corresponding to different user behavior guidance methods, including pre-acquisition probability behaviors and corresponding pre-acquisition amounts, and then based on the user behavior prediction results corresponding to different user behavior guidance methods, the expected amounts corresponding to different user behavior guidance methods are determined, further avoiding the sample unevenness problem existing in directly predicting the amount, and the predicted amount gain can be obtained through the expected amounts corresponding to different user behavior guidance methods, and the predicted amount gain can be used to improve the targeted guidance of subsequent work and improve the accuracy of operations.

[0088] The expected amount refers to the product of the pre-acquisition behavior probability and the corresponding pre-acquisition amount, reflects the convergence value of the pre-acquisition amount of a user group, and is used to indicate the predicted pre-acquisition amount value of the user group.

[0089] The first behavior guidance feature corresponds to the feature of user behavior guidance, and the second behavior guidance feature corresponds to the feature of no user behavior guidance. For example, in a marketing scenario, the first behavior guidance feature is that marketing behavior is performed, while the second behavior guidance feature is that no marketing behavior is performed.

[0090] Among them, the first pre-acquisition behavior probability and the first pre-acquisition amount under the condition of user behavior guidance are obtained through the prediction model. The first pre-acquisition behavior probability is used to characterize the possibility of the user pre-acquisition behavior under the condition of user behavior guidance, and the first pre-acquisition amount is used to characterize the pre-acquisition amount under the condition of user behavior guidance.

[0091] Correspondingly, the second pre-acquisition behavior probability and the second pre-acquisition amount in the case where no user behavior guidance is performed are obtained through the prediction model. The second pre-acquisition behavior probability is used to characterize the possibility of the user pre-acquisition behavior occurring in the case where no user behavior guidance is performed, and the second pre-acquisition amount is used to characterize the pre-acquisition amount in the case where no user behavior guidance is performed.

[0092] In an exemplary embodiment, in order to further improve the fit between the pre-acquisition behavior probability output by the prediction model and the actual pre-acquisition behavior probability, before determining the first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount, and determining the second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount, the following may also be performed:

[0093] The first pre-acquisition behavior probability is calibrated based on the probability calibration model, and the second pre-acquisition behavior probability is calibrated based on the probability calibration model.

[0094] Among them, the first pre-acquisition behavior probability and the second pre-acquisition behavior probability can be calibrated respectively using the rank-preserving regression model according to the training set and the test set. The rank-preserving regression can map the originally loosely distributed values ​​to the real conversion probability while keeping the order of the original output unchanged. Of course, the embodiments of the present disclosure can also use other probability calibration methods, which are not particularly limited.

[0095] By using the probability calibration model to calibrate the first pre-acquisition probability and the second pre-acquisition probability, the accuracy of each probability can be improved, thereby improving the accuracy of the subsequent calculation of the expected amount.

[0096] In an exemplary embodiment, determining the first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount, and determining the second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount may include:

[0097] In response to the first pre-acquisition behavior probability being lower than a preset probability threshold, determining that the first expected amount corresponding to the first pre-acquisition behavior probability is a null value;

[0098] Alternatively, in response to the second pre-acquisition behavior probability being lower than a preset probability threshold, it is determined that the second expected amount corresponding to the second pre-acquisition behavior probability is a null value.

[0099] The preset probability threshold may be determined according to the corresponding situation of the calibrated probability value and the actual conversion during the probability calibration process, so that the recall rate above the preset probability threshold is higher.

[0100] The predicted amount gain can be determined based on the first expected amount and the second expected amount. The predicted amount gain is used to characterize the increased amount of revenue when the user guidance behavior is performed compared to when the user behavior guidance is not performed, and can indicate the revenue of the user guidance investment to a certain extent. For example, under the same user guidance investment (such as telephone marketing investment), if the predicted amount gain is larger, it means that the user guidance investment is more worthwhile and can obtain greater revenue, and then further user behavior guidance can be carried out for this user group; on the contrary, if the predicted amount gain is smaller, it means that the user guidance investment has not brought considerable revenue, and then the user guidance investment can be reduced.

[0101] The following formulas (4) and (5) are used to respectively predict whether to conduct the expected amount under user behavior guidance and determine the predicted amount gain based on the expected amount:

[0102] ΔE=E(T=1)-E(T=0) (4)

[0103] E(T)=max(0,p(T)-k))*amt(T) (5)

[0104] Wherein, T is whether user behavior guidance is performed, wherein T=1 means user behavior guidance is performed, T=0 means user behavior guidance is not performed, E(T) is the predicted expected amount, p(T) is the pre-acquisition behavior probability or the calibrated pre-acquisition behavior probability, k is the preset probability threshold, amt(T) is the pre-acquisition amount corresponding to the pre-acquisition behavior, and △E is the predicted amount gain.

[0105] Referring to formulas (4) and (5), when the pre-acquisition behavior probability is lower than the preset probability threshold, it indicates that even if the user obtains the corresponding pre-acquisition amount, he may not actually perform the pre-acquisition behavior (i.e., will not convert). Therefore, the expected amount corresponding to the pre-acquisition behavior probability is determined to be a null value, thereby reducing noise and prediction errors.

[0106] It is worth noting that when the predicted amount gain △E is a negative value, this type of user can be classified as a group of people, that is, a user group that will not perform pre-acquisition behavior. The key to the implementation of the present disclosure is to analyze the predicted amount gain of the customer group with a non-negative predicted amount gain △E, so as to make reasonable user behavior guidance investment and planning based on the analysis results.

[0107] As mentioned above, during the training process of the prediction model in the marketing scenario, the behavior guidance features may include marketing features of users being marketed and users not being marketed, the pre-acquired behavior probability is the predicted order probability, and the pre-acquired amount is the predicted order amount corresponding to the execution of the order. Figure 8 A flow chart of an information prediction method in a marketing application scenario is shown, which is used to predict the order probability and order amount of a user group.

[0108] Combine the following Figure 8 The application scenario shown illustrates the information prediction method of the embodiment of the present disclosure.

[0109] First, the current user characteristics and the first marketing characteristics (telemarketing: T=1) are input into the prediction model to obtain the first order probability and the corresponding first order amount;

[0110] Secondly, the current user characteristics and the second marketing characteristics (no telephone marketing: T=0) are input into the prediction model to obtain the second order probability and the second order amount;

[0111] Furthermore, a probability calibration model is used to calibrate the first order probability and the second order probability respectively, and the calibrated first order probability is multiplied by the first order amount to obtain the first expected amount, and the calibrated second order probability is multiplied by the second order amount to obtain the second expected amount.

[0112] Finally, the difference between the first expected amount and the second expected amount is obtained to obtain the predicted amount gain.

[0113] Among them, based on the obtained predicted amount gain, it can be determined whether the user group can bring the expected amount of profit and how much expected profit it can bring when being marketed. Then, when the marketing cost is fixed, the predicted amount gain of each user group can be used to achieve precise marketing.

[0114] It should be noted that the embodiments of the present disclosure do not impose any special limitations on marketing methods, including but not limited to telephone marketing, SMS marketing, web marketing, advertising marketing, etc.

[0115] In the scenario of item recommendation, the behavior guidance features include recommending candidate items to users and not recommending candidate items to users, the pre-acquired behavior probability is the predicted order probability for the current item, and the pre-acquired amount is the predicted order amount corresponding to the execution of the order. Based on the information recommendation method of the embodiment of the present disclosure, the first expected amount in the case of recommending candidate items to users and the second expected amount in the case of not recommending candidate items to users can be obtained, and the predicted amount gain can be determined based on the first expected amount and the second expected amount.

[0116] The predicted monetary gain in the item recommendation scenario determines whether and how much expected monetary gain a user group can bring when they know the candidate items compared to when they do not know the candidate items. Based on the expected monetary gain, it is then determined which user groups need pre-recommendations and which do not, thereby achieving accurate item pre-recommendations with fixed input costs.

[0117] In summary, through the prediction scheme of the user's pre-acquisition behavior of the embodiment of the present disclosure, the pre-acquisition behavior probability and the pre-acquisition amount corresponding to the pre-acquisition behavior probability are predicted simultaneously through the multi-task model, which can avoid the sample unevenness problem existing when only predicting the pre-acquisition amount for the user, and respectively combine the current user characteristics and input them into the prediction model to obtain the user behavior prediction results corresponding to different user behavior guidance methods, including the pre-acquisition probability behavior and the corresponding pre-acquisition amount, and then determine the expected amount corresponding to different user behavior guidance methods based on the user behavior prediction results corresponding to different user behavior guidance methods, further avoiding the sample unevenness problem existing when directly predicting the amount, and the predicted amount gain can be obtained by the expected amount corresponding to different user behavior guidance methods, and the predicted amount gain can be used to improve the guidance of subsequent work and improve the accuracy of the operation; the attention feature is combined with the first feature representation and then continued to train, and the enhanced processing of the intermediate features of the model is realized based on the behavior guidance feature, the performance of the model on the behavior guidance feature is increased, and its ability to distinguish the behavior guidance feature is strengthened, thereby improving the accuracy of the model's user behavior prediction. In addition, through the joint training of multiple multi-task models, the robustness of the model is enhanced and the maintenance cost is reduced.

[0118] In an exemplary embodiment of the present disclosure, a model training device 900 is also provided. Fig. 9 , the model training device 900 includes:

[0119] A first processing module 910 is used to input the user sample feature into a feature sharing network in the multi-task model to obtain a first feature representation, where the user sample feature is obtained according to the user feature and the corresponding behavior guidance feature;

[0120] A second processing module 920 is used to process the behavior guidance feature through the attention structure in the multi-task model, and determine a second feature representation by combining the obtained attention feature and the first feature representation;

[0121] The model parameter adjustment module 930 is used to use the second feature representation as the input of the dual-tower model structure in the multi-task model, predict the pre-acquisition behavior probability and the corresponding pre-acquisition amount through the dual-tower model structure, and adjust the multi-task model based on the pre-acquisition behavior probability and the pre-acquisition amount in combination with the sample standard label.

[0122] In an exemplary embodiment of the present disclosure, the user characteristics include user attribute characteristics and user behavior characteristics, and the first processing module 910 further includes:

[0123] The feature determination module is used to combine user attribute features, user behavior features and behavior guidance features to obtain user sample features.

[0124] In an exemplary embodiment of the present disclosure, the dual-tower model structure includes a first prediction model and a second prediction model; the model parameter adjustment module 930 also includes:

[0125] A first processing unit is used to perform prediction processing on the second feature representation using the first prediction model to obtain the pre-acquisition behavior probability corresponding to the user sample feature; a second processing unit is used to perform prediction processing on the second feature representation according to the second prediction model based on the pre-acquisition behavior probability to obtain the pre-acquisition amount; a parameter updating unit is used to determine a first loss function according to the pre-acquisition behavior probability and the corresponding standard behavior label, and to determine a second loss function according to the pre-acquisition amount and the corresponding standard amount label, so as to adjust the parameters in the multi-task model in combination with the first loss function and the second loss function.

[0126] In an exemplary embodiment of the present disclosure, the parameter updating unit is configured to: if the pre-acquired amount is a non-null value, adjust the parameters in the first prediction model and the second prediction model in combination with the first loss function and the second loss function; if the pre-acquired amount is a null value, adjust the parameters in the first prediction model according to the first loss function.

[0127] In an exemplary embodiment of the present disclosure, the model training device 900 further includes:

[0128] The acquisition module is used to divide the historical pre-acquisition amounts of sample users into multiple groups; and determine the standard amount label corresponding to the group according to the historical pre-acquisition amounts of sample users in each group.

[0129] Since the various functional modules of the model training device of the exemplary embodiment of the present disclosure are the same as those in the inventive implementation of the above-mentioned model training method, they will not be described in detail here.

[0130] In an exemplary embodiment of the present disclosure, an information prediction device 1000 is also provided. Fig.10 As shown, the information prediction device 1000 may include:

[0131] The first prediction module 1010 is used to use the current user feature and the first behavior guiding feature as inputs of the prediction model to predict a first pre-acquisition behavior probability and a first pre-acquisition amount corresponding to the first pre-acquisition behavior probability;

[0132] The second prediction module 1020 is used to input the current user feature and the second behavior guidance feature into the prediction model to predict a second pre-acquisition behavior probability and a second pre-acquisition amount corresponding to the second pre-acquisition behavior probability, wherein the first behavior guidance feature and the second behavior guidance feature respectively correspond to whether to perform user behavior guidance;

[0133] A gain prediction module 1030 is used to determine a first expected amount based on a first pre-acquisition behavior probability and a first pre-acquisition amount, determine a second expected amount based on a second pre-acquisition behavior probability and a second pre-acquisition amount, and determine a predicted amount gain based on the first expected amount and the second expected amount;

[0134] Among them, the prediction model is obtained after training the multi-task model according to the above model training method.

[0135] In an exemplary embodiment of the present disclosure, the information prediction device 1000 further includes:

[0136] A probability calibration module is used to calibrate the first pre-acquisition behavior probability based on a probability calibration model, and to calibrate the second pre-acquisition behavior probability based on the probability calibration model.

[0137] In an exemplary embodiment of the present disclosure, the gain prediction module 1030 includes:

[0138] a first determining unit, configured to determine, in response to the first pre-acquisition behavior probability being lower than a preset probability threshold, that a first expected amount corresponding to the first pre-acquisition behavior probability is a null value;

[0139] Alternatively, the second determining unit is configured to determine, in response to the second pre-acquisition behavior probability being lower than the preset probability threshold, that the second expected amount corresponding to the second pre-acquisition behavior probability is a null value.

[0140] Since the functional modules of the information prediction device according to the exemplary embodiment of the present disclosure are the same as those in the above-mentioned information prediction method, they will not be described in detail here.

[0141] It should be noted that although several modules or units of the model training device and the information prediction device are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0142] In addition, in the exemplary embodiment of the present disclosure, a computer storage medium capable of implementing the above method is also provided. A program product capable of implementing the above method of this specification is stored thereon. In some possible embodiments, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to perform the steps according to various exemplary embodiments of the present disclosure described in the above "Exemplary Method" section of this specification.

[0143] The exemplary embodiments of the present disclosure also provide a program product for implementing the above method, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0144] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0145] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0146] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0147] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0148] In addition, in the exemplary embodiments of the present disclosure, an electronic device capable of implementing the above method is also provided. It will be appreciated by those skilled in the art that various aspects of the present disclosure may be implemented as a system, method, or program product. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: a complete hardware embodiment, a complete software embodiment (including firmware, microcode, etc.), or an embodiment combining hardware and software aspects, which may be collectively referred to herein as a "circuit", "module", or "system".

[0149] Refer to the following Fig.11 1100 according to this embodiment of the present disclosure is described. Fig.11 The electronic device 1100 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0150] like Fig.11 As shown, the electronic device 1100 is in the form of a general computing device. The components of the electronic device 1100 may include, but are not limited to: the at least one processing unit 1110, the at least one storage unit 1120, a bus 1130 connecting different system components (including the storage unit 1120 and the processing unit 1110), and a display unit 1140.

[0151] The storage unit stores program codes, which can be executed by the processing unit 1110, so that the processing unit 1110 executes the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification.

[0152] The storage unit 1120 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 11201 and / or a cache storage unit 11202 , and may further include a read-only storage unit (ROM) 11203 .

[0153] The storage unit 1120 may also include a program / utility 11204 having a set (at least one) of program modules 11205, such program modules 11205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0154] Bus 1130 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0155] The electronic device 1100 may also communicate with one or more external devices 1200 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1100, and / or communicate with any device that enables the electronic device 1100 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1150. Furthermore, the electronic device 1100 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 1160. As shown, the network adapter 1160 communicates with other modules of the electronic device 1100 via a bus 1130. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1100, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0156] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.

[0157] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0158] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. A model training method, It is characterized in that include: Inputting the user sample features into a feature sharing network in the multi-task model to obtain a first feature representation, wherein the user sample features are obtained according to the user features and the corresponding behavior guidance features; Processing the behavior guidance feature through the attention structure in the multi-task model, and determining a second feature representation by combining the obtained attention feature and the first feature representation; The second feature representation is used as the input of the dual-tower model structure in the multi-task model. The pre-acquisition behavior probability and the corresponding pre-acquisition amount are predicted respectively through the dual-tower model structure, and the multi-task model is adjusted based on the pre-acquisition behavior probability and the pre-acquisition amount in combination with the sample standard label.

2. The method according to claim 1, It is characterized in that The user characteristics include user attribute characteristics and user behavior characteristics. The user sample characteristics obtained according to the user characteristics and the corresponding behavior guidance characteristics include: The user attribute features, the user behavior features and the behavior guidance features are concatenated to obtain the user sample features.

3. The method according to claim 1, It is characterized in that The dual-tower model structure includes a first prediction model and a second prediction model; the dual-tower model structure is used to respectively predict the pre-acquisition behavior probability and the corresponding pre-acquisition amount, and based on the pre-acquisition behavior probability and the pre-acquisition amount, the multi-task model is adjusted in combination with the sample standard label, including: Using the first prediction model to perform prediction processing on the second feature representation to obtain the pre-acquisition behavior probability corresponding to the user sample feature; Based on the pre-acquisition behavior probability, performing prediction processing on the second feature representation according to the second prediction model to obtain the pre-acquisition amount; A first loss function is determined based on the pre-acquired behavior probability and the corresponding standard behavior label, and a second loss function is determined based on the pre-acquired amount and the corresponding standard amount label, so as to adjust the parameters in the multi-task model in combination with the first loss function and the second loss function.

4. The method according to claim 3, It is characterized in that The determining of a first loss function according to the pre-acquired behavior probability and the corresponding standard behavior label, and determining a second loss function according to the pre-acquired amount and the corresponding standard amount label, so as to adjust the parameters in the multi-task model in combination with the first loss function and the second loss function, includes: If the pre-acquired amount is a non-null value, adjusting parameters in the first prediction model and the second prediction model in combination with the first loss function and the second loss function; If the pre-acquired amount is a null value, the parameters in the first prediction model are adjusted according to the first loss function.

5. The method according to claim 3, It is characterized in that Before determining the second loss function according to the pre-acquired amount and the corresponding standard amount label, the method further includes: Divide the historical pre-acquisition amounts of sample users into multiple groups; The standard amount label corresponding to each group is determined according to the historical pre-acquired amounts of sample users in the group.

6. An information prediction method, It is characterized in that include: Using the current user feature and the first behavior guiding feature as inputs of the prediction model, predicting a first pre-acquisition behavior probability and a first pre-acquisition amount corresponding to the first pre-acquisition behavior probability; Inputting the current user feature and the second behavior guidance feature into the prediction model, predicting a second pre-acquisition behavior probability and a second pre-acquisition amount corresponding to the second pre-acquisition behavior probability, the first behavior guidance feature and the second behavior guidance feature respectively corresponding to whether to perform user behavior guidance; determining a first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount, determining a second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount, and determining a predicted amount gain based on the first expected amount and the second expected amount; The prediction model is obtained by training the multi-task model according to the model training method described in any one of claims 1 to 5.

7. The method according to claim 6, It is characterized in that Before determining the first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount, and determining the second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount, the method further includes: The first pre-acquisition behavior probability is calibrated based on a probability calibration model, and the second pre-acquisition behavior probability is calibrated based on the probability calibration model.

8. The method according to claim 6, It is characterized in that The determining of the first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount, and determining the second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount, includes: In response to the first pre-acquisition behavior probability being lower than a preset probability threshold, determining that a first expected amount corresponding to the first pre-acquisition behavior probability is a null value; Alternatively, in response to the second pre-acquisition behavior probability being lower than the preset probability threshold, it is determined that the second expected amount corresponding to the second pre-acquisition behavior probability is a null value.

9. A model training device, It is characterized in that include: A first processing module, configured to input user sample features into a feature sharing network in a multi-task model to obtain a first feature representation, wherein the user sample features are obtained according to the user features and corresponding behavior guidance features; A second processing module, configured to process the behavior guidance feature through the attention structure in the multi-task model, and determine a second feature representation by combining the obtained attention feature and the first feature representation; A model parameter adjustment module is used to use the second feature representation as an input of the dual-tower model structure in the multi-task model, predict the pre-acquisition behavior probability and the corresponding pre-acquisition amount through the dual-tower model structure, and adjust the multi-task model based on the pre-acquisition behavior probability and the pre-acquisition amount in combination with the sample standard label.

10. An information prediction device, It is characterized in that include: A first prediction module, configured to use the current user feature and the first behavior guiding feature as inputs of a prediction model, and predict a first pre-acquisition behavior probability and a first pre-acquisition amount corresponding to the first pre-acquisition behavior probability; a second prediction module, configured to input the current user feature and the second behavior guidance feature into the prediction model, and predict a second pre-acquisition behavior probability and a second pre-acquisition amount corresponding to the second pre-acquisition behavior probability, wherein the first behavior guidance feature and the second behavior guidance feature respectively correspond to whether to perform user behavior guidance; a gain prediction module, configured to determine a first expected amount based on the first pre-acquisition behavior probability and the first pre-acquisition amount, determine a second expected amount based on the second pre-acquisition behavior probability and the second pre-acquisition amount, and determine a predicted amount gain based on the first expected amount and the second expected amount; The prediction model is obtained by training a multi-task model according to the method described in any one of claims 1 to 5.

11. A computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the method according to any one of claims 1 to 8 when executed by a processor.

12. An electronic device, It is characterized in that include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 8 by executing the executable instructions.

Images