Method, apparatus, storage medium and electronic device for training result prediction model

Abstract

The embodiments of this specification disclose a method, device, storage medium, and electronic device for training a result prediction model. The method first trains a fusion model based on first-sample training data, then obtains counterfactual results based on the trained fusion model, constructs unbiased augmented signals based on the first-sample training data and the outputs of the trained fusion model for each decision item, and then trains the result prediction model based on the unbiased augmented signals. Thus, by combining a regression model and a propensity score model for multi-task representation learning and fusing multiple prediction models to share parameters to reduce variance, it can effectively solve the data bias problem and improve the training efficiency and prediction accuracy.

Description

Technical Field

[0001] The present invention relates to computer technology, and in particular, to a method, device, storage medium, and electronic device for training a result prediction model. Background Art

[0002] In actual application scenarios, decision-making problems often exist. For example, scenarios such as outbound call timing selection and identity verification recommendation are essentially decision-making problems, and all require predicting results based on customer characteristics. The above decision-making problems belong to causal inference. The difficulty in solving such decision-making problems in the prior art lies in the problem of selection bias. On the one hand, there is sample bias. Taking the outbound call timing selection scenario as an example, usually what is expected to be predicted is the answering rate of each type of customer at each time point. However, the training data collected often only has the results of this type of user answering the phone at a specific time point, and the labels for outbound calls at other time points are missing. On the other hand, there is feature bias. Still taking the outbound call timing selection scenario as an example, the customer feature distributions for outbound calls at each time point based on historical policies are different. This distribution difference will cause the model to lack the ability to predict counterfactual results. Summary of the Invention

[0003] The purpose of the embodiments of this specification is to provide a method, device, storage medium, and electronic device for training a result prediction model.

[0004] The embodiments of this specification provide a method for training a result prediction model, which can effectively solve the data bias problem and improve the training efficiency and prediction accuracy. The method includes:

[0005] Obtain first sample training data, where the first sample training data includes feature data of a sample object, a first decision item, and the factual result of the sample object regarding the first decision item;

[0006] Train a fusion model based on the first sample training data to obtain a trained fusion model, where the fusion model includes a regression model and a propensity score model;

[0007] Input a second decision item and the feature data into the trained fusion model to obtain a counterfactual result regarding the second decision item output by the regression model and a propensity score result output by the propensity score model;

[0008] Construct an unbiased augmented signal according to the first sample training data and the outputs of the trained fusion model for each decision item;

[0009] Construct second sample training data, where the second sample training data includes a third decision item and the feature data, and the third decision item includes the first decision item and the second decision item;

[0010] Train the result prediction model according to the second sample training data and the unbiased augmentation signal to obtain a trained result prediction model.

[0011] Further, training the fusion model based on the first sample training data to obtain a trained fusion model includes:

[0012] Input the first decision item into the regression model in the fusion model to obtain the expected result corresponding to the first decision item output by the regression model, and construct a first loss function based on the expected result and the observation result;

[0013] Input the feature data into the propensity score model in the fusion model to obtain the propensity score result corresponding to the sample object output by the propensity score model, and construct a second loss function based on the propensity score result and the first decision item;

[0014] Train the fusion model according to the first loss function and the second loss function to obtain a trained fusion model.

[0015] Further, training the fusion model according to the first loss function and the second loss function to obtain a trained fusion model includes:

[0016] Construct a third loss function according to the first loss function, the second loss function and the balance parameter, and train the fusion model according to the third loss function;

[0017] Adjust the balance parameter to make the fusion model converge to obtain a trained fusion model.

[0018] Further, the propensity score model includes a first embedding layer and a first linear layer, the regression model includes a second embedding layer and a second linear layer, and the input of the second linear layer includes the second vector output by the second embedding layer and the first vector output by the first embedding layer.

[0019] Further, training the result prediction model according to the second sample training data and the unbiased augmentation signal to obtain a trained result prediction model includes:

[0020] Input the second sample training data into the result prediction model to obtain the target expected result of the sample object regarding the third decision item output by the result prediction model, and train the result prediction model through a fourth loss function constructed based on the target expected result and the unbiased augmentation signal to obtain a trained result prediction model.

[0021] Further, the result prediction model includes a third embedding layer, a fourth embedding layer, and a third linear layer. The input of the third embedding layer is the feature data, the input of the fourth embedding layer is the third decision item, and the input of the third linear layer includes the third vector output by the third embedding layer and the fourth vector output by the fourth embedding layer.

[0022] An embodiment of this specification also provides a method for result prediction, including:

[0023] Obtaining feature data corresponding to a target object and a target decision item;

[0024] Inputting the feature data and the target decision item into a trained result prediction model to obtain a prediction result corresponding to the target object, where the result prediction model is trained based on the method for training a result prediction model described in the embodiment of this specification.

[0025] An embodiment of this specification also provides a device for training a result prediction model, including:

[0026] A first obtaining module, configured to obtain first sample training data, where the first sample training data includes feature data of a sample object, a first decision item, and a factual result of the sample object regarding the first decision item;

[0027] A first training module, configured to train a fusion model based on the first sample training data to obtain a trained fusion model, where the fusion model includes a regression model and a propensity score model;

[0028] A second obtaining module, configured to input a second decision item and the feature data into the trained fusion model to obtain a counterfactual result regarding the second decision item output by the regression model and a propensity score result output by the propensity score model;

[0029] A first construction module, configured to construct an unbiased augmented signal according to the first sample training data and the output of the trained fusion model for each decision item;

[0030] A second construction module, configured to construct second sample training data, where the second sample training data includes a third decision item and the feature data, and the third decision item includes the first decision item and the second decision item;

[0031] A second training module, configured to train a result prediction model according to the second sample training data and the unbiased augmented signal to obtain a trained result prediction model.

[0032] An embodiment of this specification also provides a device for result prediction, including:

[0033] A third acquisition module, configured to acquire feature data corresponding to a target object and a target decision item;

[0034] A fourth acquisition module, configured to input the feature data and the target decision item into a trained result prediction model to obtain a prediction result corresponding to the target object, where the result prediction model is trained based on the method for training a result prediction model described in this embodiment of the specification.

[0035] This embodiment of the specification further provides a storage medium storing a computer program, and the computer program is adapted to be loaded and executed by a processor to perform the steps of the above method.

[0036] This embodiment of the specification further provides an electronic device, including: a processor and a memory; wherein, the memory stores a computer program, and the computer program is adapted to be loaded and executed by the processor to perform the steps of the above method.

[0037] This embodiment of the specification further provides a computer program product, on which at least one instruction is stored, and characterized in that when the at least one instruction is executed by a processor, the steps of the above method are implemented.

[0038] In this embodiment of the specification, by combining a regression model and a propensity score model into a fusion model, first training the fusion model based on the observed first sample training data, then obtaining a counterfactual result based on the trained fusion model, and constructing an unbiased augmented signal according to the first sample training data and the output of the trained fusion model for each decision item, and further training a result prediction model based on the unbiased augmented signal. Thus, by combining a regression model and a propensity score model for multi-task representation learning, fusing multiple prediction models to share parameters to reduce variance, it can effectively solve the data bias problem and improve the training efficiency and prediction accuracy. Description of the Drawings

[0039] Figure 1 It is a schematic flowchart of a method for training a result prediction model provided by this embodiment of the specification.

[0040] Figure 2 It is a schematic framework diagram of an example for training a result prediction model provided by this embodiment of the specification.

[0041] Figure 3 It is a schematic flowchart of a method for result prediction provided by this embodiment of the specification.

[0042] Figure 4 It is a schematic structural diagram of a device for training a result prediction model provided by this embodiment of the specification.

[0043] Figure 5 This is a schematic structural diagram of a device for result prediction provided by an embodiment of this specification.

[0044] Figure 6 This is a schematic structural diagram of an electronic device provided by an embodiment of this specification. Detailed implementation manners

[0045] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only a part rather than all of the embodiments of this specification. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this specification.

[0046] Please refer to Figure 1 , which is a schematic flowchart of a method for training a result prediction model provided by an embodiment of this specification. In the embodiment of this specification, the method for training a result prediction model is applied to a device for training a result prediction model (hereinafter simply referred to as the "model training device") or an electronic device configured with the model training device. The following will elaborate in detail on the Figure 1 shown flowchart. The method for training a result prediction model may specifically include the following steps:

[0047] S102. Obtain first sample training data, where the first sample training data includes feature data of a sample object, a first decision item, and the factual result of the sample object with respect to the first decision item.

[0048] In some embodiments, the first sample training data is constructed based on observed factual data. In some embodiments, the sample object is a user, and the feature data of the sample object includes user feature data corresponding to the user. For example, in the scenario of outbound call timing selection, the feature data of the sample object includes customer feature data collected at each time point for outbound calls. In some embodiments, the first decision item (which may also become a "treatment group" in the context of the decision item) is each observed treatment group or decision factor. For example, in the scenario of outbound call timing selection, the first decision item can be the time point of making a call to the user as observed. Another example is that in the scenario of identity verification recommendation, the first decision item can be the identity verification method recommended, such as face, fingerprint, etc.; the factual result of the sample object with respect to the first decision item is the observed result. For example, in the scenario of outbound call timing selection, the factual result with respect to the first decision item can be the call answering result observed at the time point represented by the first decision item when making a call to the user. Another example is that in the scenario of identity verification recommendation, the factual result with respect to the first decision item can be the identity verification result observed under the identity verification method represented by the first decision item. In some embodiments, the first sample training data includes multiple samples, and each sample includes an associated first decision item, the feature data corresponding to the first decision item, and the factual result corresponding to the first decision item. During the subsequent training process of the fusion model, the label corresponding to the first decision item as an input is the factual result corresponding to the first decision item, and the label corresponding to the feature data as an input is the first decision item corresponding to the feature data. The specific training process will be described in detail in the subsequent embodiments and will not be elaborated here.

[0049] In some embodiments, the observation result can be obtained through observation over a period of time, and the first sample training data is constructed based on the observation result. In some embodiments, the first sample training data provided by other devices can be obtained. It should be noted that the above embodiments for obtaining the first sample training data are only examples and not limitations to this application. Those skilled in the art should understand that any embodiment for obtaining the first sample training data should be included within the protection scope of this application.

[0050] S104. Train the fusion model based on the first sample training data to obtain a trained fusion model, where the fusion model includes a regression model and a propensity score model.

[0051] In some embodiments, the fusion model is obtained by combining a regression model and a propensity score model, and the regression model and the propensity score model have shared parameters. In some embodiments, each sample in the first sample training data is input into the fusion model for training. If it does not converge, the model parameters are adjusted and training continues. If it converges, a trained fusion model is obtained.

[0052] In some embodiments, the propensity score model includes a first embedding layer and a first linear layer, and the regression model includes a second embedding layer and a second linear layer. The input of the second linear layer includes a second vector output by the second embedding layer and a first vector output by the first embedding layer. In some embodiments, during the training process, a sample in the first sample training data is input into the fusion model. The input of the second embedding layer is the first decision item in this sample (i.e., the input of the regression model is the first decision item), and the output of the second linear layer is the prediction result for the first decision item, that is, the output of the regression model. The input of the first embedding layer is the feature data in this sample (i.e., the input of the propensity score model is the feature data), and the output of the first linear layer is the propensity score for this feature data, that is, the output of the propensity score model. In some embodiments, the propensity score model further includes a neural network provided at the front end of the first embedding layer. For example, a KAN (Kolmogorov - Arnold Networks, representing theorem network) and an MLP (Multilayer Perceptron) are sequentially provided at the front end of the first embedding layer. Another example is that a fully connected layer is provided at the front end of the first embedding layer. In some embodiments, the first linear layer and the second linear layer can also be replaced with other neural network structures. For example, they can be replaced with fully connected layers. The present application does not limit the specific network structure in the fusion model. Those skilled in the art should understand that any existing or future possible network structure for implementing the regression model and the propensity score model can be used to form the fusion model described in the present application.

[0053] S106. Input the second decision item and the feature data into the trained fusion model to obtain the counterfactual result regarding the second decision item output by the regression model and the propensity score result output by the propensity score model.

[0054] In some embodiments, the second decision item is different from the first decision item. In some embodiments, the second decision item is an unobserved treatment group or decision factor. For example, in the scenario of outbound call timing, if the call time points are divided into 13 time points from 9 o'clock to 19 o'clock by hour, and the observed call time points are 9 o'clock, 10 o'clock, 12 o'clock, 14 o'clock, 15 o'clock, 17 o'clock, 19 o'clock, then the observed above time points are the first decision items, and the unobserved 11 o'clock, 13 o'clock, 16 o'clock, 18 o'clock are the second decision items. Then the result predicted based on the second decision item is the counterfactual result.

[0055] In some embodiments, based on the trained fusion model, the counterfactual results corresponding to all the second decision items and the associated propensity score results can be obtained. Combining with the propensity score results corresponding to the first decision item during the training process of the fusion model, that is, at this time, all the counterfactual results and the propensity score results corresponding to all the decision items have been obtained, and thus the subsequent operation of constructing the unbiased augmented signal can be performed.

[0056] S108. Construct an unbiased augmented signal according to the first sample training data and the output of the obtained fusion model for each decision item.

[0057] In some embodiments, the output for each decision item includes the output results of the regression model and the propensity score model in step S104 and the output results of the regression model and the propensity score model in step S106. In some embodiments, first, an augmented signal is constructed based on the first sample training data and the output of the obtained fusion model for each decision item, then the features are regressed using the constructed augmented signal, and then unbiasedness derivation is performed to obtain the unbiased augmented signal.

[0058] As an example, the augmented signal is constructed as follows:

[0059]

[0060] where t represents the t-th decision item, x represents the feature data of the sample object, T represents the input decision item, represents the augmented signal constructed based on the estimated value of the regression model and the estimated value of the propensity model, represents the expected result of the regression model for the t-th decision item (i.e., the output of the regression model for the t-th decision item), represents the propensity score result of the propensity score model for the t-th decision item, the function takes the value of 1 when T = t and 0 when T ≠ t, Y represents the label corresponding to the t-th decision item (i.e., the factual result); the unbiased augmented signal can be obtained by performing unbiasedness derivation on the above augmented signal.

[0061] S110. Construct the second sample training data, where the second sample training data includes the third decision item and the feature data, and the third decision item includes the first decision item and the second decision item. Wherein, the second sample training data is used to train the result prediction model.

[0062] S112. Train the result prediction model according to the second sample training data and the unbiased augmented signal to obtain the trained result prediction model.

[0063] In some embodiments, the unbiased augmented signal is used as the label of the second sample training data, and a supervised algorithm is used to train the result prediction model. In some embodiments, the result prediction model includes a third embedding layer, a fourth embedding layer, and a third linear layer. The input of the third embedding layer is the feature data, the input of the fourth embedding layer is the third decision item, and the input of the third linear layer includes the third vector output by the third embedding layer and the fourth vector output by the fourth embedding layer. In some embodiments, the result prediction model further includes other neural network structures arranged at the front end of the third embedding layer, and this part of the neural network structure is the same as or similar to the neural network structure arranged at the front end of the propensity score model in the fusion model. In some embodiments, the third linear layer can also be replaced by other neural network structures, such as a fully connected layer.

[0064] In some embodiments, training the fusion model based on the first sample training data to obtain a trained fusion model includes: inputting the first decision item into the regression model in the fusion model to obtain the expected result corresponding to the first decision item output by the regression model, and constructing a first loss function based on the expected result and the observed result; inputting the feature data into the propensity score model in the fusion model to obtain the propensity score result corresponding to the sample object output by the propensity score model, and constructing a second loss function based on the propensity score result and the first decision item; training the fusion model according to the first loss function and the second loss function to obtain a trained fusion model. In some embodiments, when inputting the first decision item, the feature data associated with the first decision item is input into the propensity score model in the fusion model. The factual result corresponding to the first decision item is used as the label to be compared with the output of the regression model, and the first decision item is used as the label to be compared with the output of the propensity score model. Thus, the first loss function corresponding to the regression model branch and the second loss function corresponding to the propensity score model can be respectively constructed to obtain a trained fusion model.

[0065] In some embodiments, training the fusion model according to the first loss function and the second loss function to obtain a trained fusion model includes: constructing a third loss function according to the first loss function, the second loss function, and a balance parameter, and training the fusion model according to the third loss function; adjusting the balance parameter to converge the fusion model to obtain a trained fusion model. In some embodiments, the balance parameter is preset, and the fusion model can be trained respectively based on a plurality of preset balance parameters to obtain a converged fusion model. In some embodiments, the balance parameter can be dynamically adjusted proportionally during the training process to obtain a converged fusion model. As an example, set the balance parameter λ (λ is a value greater than 0) and initialize the network weights, and train the fusion model based on the input data (x i , y i , t i ) of the first sample training data, where x i represents the feature data of the i-th sample, y i represents the factual result of the i-th sample, and t i represents the first decision item corresponding to the i-th sample; if not converged, execute the following algorithm: for the regression model branch in the fusion model, calculate the first loss function corresponding to the regression model branch:

[0066]

[0067] where, represents the first loss function, represents the form of the loss function, , , represent the neural network functions in the regression model, represents the feature data corresponding to the i-th sample, represents the first decision item corresponding to the i-th sample, represents the factual result corresponding to the i-th sample; for the propensity score model branch in the fusion model, calculate the second loss function corresponding to the propensity score model branch:

[0068]

[0069] where, represents the second loss function, represents the form of the loss function, , represent the neural network functions in the propensity score model, represents the feature data corresponding to the i-th sample, represents the first decision item corresponding to the i-th sample; then based on to update the network weights of the fusion model. And so on until the fusion model converges.

[0070] In some embodiments, training the result prediction model according to the second sample training data and the unbiased augmentation signal to obtain a trained result prediction model includes: inputting the second sample training data into the result prediction model to obtain the target expected result of the sample object regarding the third decision item output by the result prediction model, and training the result prediction model through a fourth loss function constructed based on the target expected result and the unbiased augmentation signal to obtain a trained result prediction model. In some embodiments, the unbiased augmentation signal is used as a label for supervised training to obtain a converged result prediction model.

[0071] This application discovers that the existing DRL (Doubly Robust Learner) as a means to solve selection bias only supports the sklearn (a machine learning library) model framework, does not support heterogeneous data for representation learning, and a prediction model needs to be established for each decision item, resulting in complex deployment due to the large number of inputs to the prediction model in practical applications. The solution of the embodiments of this specification is creatively extended based on the idea of traditional DRL. By combining a regression model and a propensity score model for multi-task representation learning, multiple prediction models can be fused to share parameters and reduce variance. Through experiments, this solution has significant advantages over existing technologies (such as Baseline, IPW, etc.) in both offline and online tests and can more effectively solve data bias.

[0072] According to the solution of the embodiments of this specification, by combining a regression model and a propensity score model into a fusion model, first training the fusion model based on the observed first sample training data, then obtaining counterfactual results based on the trained fusion model, and constructing an unbiased augmentation signal according to the first sample training data and the outputs of the trained fusion model for each decision item, and then training the result prediction model based on the unbiased augmentation signal. Thus, through combining a regression model and a propensity score model for multi-task representation learning, fusing multiple prediction models to share parameters to reduce variance, it can effectively solve the data bias problem, facilitate deployment, and improve the model training efficiency and model prediction accuracy.

[0073] Figure 2 A schematic diagram of a framework for training a result prediction model provided by an example of the embodiments of this specification. As Figure 2As shown in the figure, the fusion model includes a regression model branch and a propensity score model branch. The input of the regression model branch is denoted as T. The regression model branch includes a second embedding layer and a second linear layer. During the training process, based on the output of the regression model and the label Y (i.e., the factual result corresponding to the input T during the training process), the loss Loss1 of the regression model branch can be obtained. The input of the propensity score model branch is denoted as X. The propensity score model branch includes KAN, MLP, a first embedding layer, and a first linear layer. Based on the output of the propensity score model branch and the label T (i.e., the decision item corresponding to the input X during the training process), the loss Loss2 of the propensity score model branch can be obtained. The above Loss1 and Loss2 are used to train the fusion model. After that, all counterfactual results and associated propensity score results can be obtained based on the trained fusion model. Based on the outputs of the two branches of the fusion model in the above steps, an unbiased augmented signal can be constructed as the label for training the result prediction model. Then, the result prediction model can be trained, such as Figure 2 As shown in the figure, the input of the result prediction model includes T and X. The result prediction model includes a third embedding layer, a fourth embedding layer, a third linear layer, and KAN and MLP in front of the third embedding layer. Based on the output of the result prediction model and the unbiased augmented signal, the loss Loss can be obtained. The parameters of the result prediction model can be adjusted based on Loss to train a converged result prediction model. Such as Figure 2 As can be seen from the arrow pointing from the embedding layer to the second linear layer in the figure, the regression model branch and the propensity score model branch share the feature data X input to the fusion model.

[0074] Figure 3 This is a schematic flowchart of a method for result prediction provided by an embodiment of this specification. The method includes the following steps:

[0075] S202, Obtain the feature data corresponding to the target object and the target decision item.

[0076] For example, in the timing scenario of outbound calls, obtain the feature data corresponding to the user to be called and the target decision item, where the target decision item is used to indicate the time node for making a call. For another example, in the scenario of identity verification and recommendation, obtain the feature data corresponding to the user to be verified and the target decision item, where the target decision item is used to indicate the identity verification method, such as face, fingerprint, gesture, etc.

[0077] S204, Input the feature data and the target decision item into the trained result prediction model to obtain the prediction result corresponding to the target object, where the result prediction model is trained based on the method for training the result prediction model described in the embodiment of this specification.

[0078] For example, in the scenario of timing for outbound calls, in step S202, the feature data corresponding to the user to be called and the target decision item are obtained. The target decision item is used to indicate the time node for making a call. In step S204, the feature data and the target decision item are input into the trained result prediction model to obtain the prediction result of making a call at this time node, which is also the result predicted for whether the user will answer the call at this time node. Based on the result prediction model, various metrics corresponding to the target object can be further evaluated. For example, in the scenario of timing for outbound calls, by predicting the answer results of calling the user at each time node, the answer rate of the user can be evaluated.

[0079] Figure 4 FIG. is a schematic structural diagram of an apparatus for training a result prediction model provided by an embodiment of the present specification. The apparatus for training a result prediction model (hereinafter simply referred to as "model training apparatus 1") can be implemented as all or part of an electronic device through software, hardware, or a combination of both. According to some embodiments, the model training apparatus 1 includes a first acquisition module 11, a first training module 12, a second acquisition module 13, a first construction module 14, a second construction module 15, and a second training module 16.

[0080] The first acquisition module 11 is configured to acquire first sample training data, where the first sample training data includes the feature data of the sample object, the first decision item, and the factual result of the sample object with respect to the first decision item.

[0081] The first training module 12 is configured to train the fusion model based on the first sample training data to obtain a trained fusion model, where the fusion model includes a regression model and a propensity score model.

[0082] The second acquisition module 13 is configured to input the second decision item and the feature data into the trained fusion model to obtain the counterfactual result of the regression model with respect to the second decision item and the propensity score result output by the propensity score model.

[0083] The first construction module 14 is configured to construct an unbiased augmented signal according to the first sample training data and the output of the trained fusion model for each decision item.

[0084] The second construction module 15 is configured to construct second sample training data, where the second sample training data includes a third decision item and the feature data, and the third decision item includes the first decision item and the second decision item.

[0085] The second training module 16 is configured to train the result prediction model according to the second sample training data and the unbiased augmented signal to obtain a trained result prediction model.

[0086] In some embodiments, the first training module 12 is configured to:

[0087] Input the first decision item into the regression model in the fusion model to obtain the expected result corresponding to the first decision item output by the regression model, and construct a first loss function based on the expected result and the observation result;

[0088] Input the feature data into the propensity score model in the fusion model to obtain the propensity score result corresponding to the sample object output by the propensity score model, and construct a second loss function based on the propensity score result and the first decision item;

[0089] Train the fusion model according to the first loss function and the second loss function to obtain a trained fusion model.

[0090] In some embodiments, the training the fusion model according to the first loss function and the second loss function to obtain a trained fusion model includes:

[0091] Construct a third loss function according to the first loss function, the second loss function and a balance parameter, and train the fusion model according to the third loss function;

[0092] Adjust the balance parameter to make the fusion model converge to obtain a trained fusion model.

[0093] In some embodiments, the propensity score model includes a first embedding layer and a first linear layer, the regression model includes a second embedding layer and a second linear layer, and the input of the second linear layer includes a second vector output by the second embedding layer and a first vector output by the first embedding layer.

[0094] In some embodiments, the second training module 16 is configured to:

[0095] Input the second sample training data into the result prediction model to obtain the target expected result of the sample object regarding the third decision item output by the result prediction model, and train the result prediction model through a fourth loss function constructed based on the target expected result and the unbiased augmentation signal to obtain a trained result prediction model.

[0096] In some embodiments, the result prediction model includes a third embedding layer, a fourth embedding layer and a third linear layer, the input of the third embedding layer is the feature data, the input of the fourth embedding layer is the third decision item, and the input of the third linear layer includes a third vector output by the third embedding layer and a fourth vector output by the fourth embedding layer.

[0097] Figure 5 The following is a schematic structural diagram of a device for result prediction provided by an embodiment of this specification. The device for result prediction (hereinafter simply referred to as "result prediction device 2") can be implemented as all or part of an electronic device through software, hardware, or a combination of both. According to some embodiments, the result prediction device 2 includes a third acquisition module 21 and a fourth acquisition module 22.

[0098] The third acquisition module 21 is configured to acquire feature data corresponding to a target object and a target decision item;

[0099] The fourth acquisition module 22 is configured to input the feature data and the target decision item into a trained result prediction model to obtain a prediction result corresponding to the target object, where the result prediction model is trained based on the method for training a result prediction model described in the embodiments of this specification.

[0100] The above device embodiments correspond to the method embodiments. For specific descriptions, reference can be made to the descriptions in the method embodiment part, which will not be elaborated here. The device embodiments are obtained based on the corresponding method embodiments and have the same technical effects as the corresponding method embodiments. For specific descriptions, reference can be made to the corresponding method embodiments.

[0101] An embodiment of this specification further provides a computer storage medium, which can store multiple instructions, and the instructions are suitable for being loaded and executed by a processor to implement the method described in the embodiments of this specification.

[0102] An embodiment of this specification further provides a computer program product, which stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the method described in the embodiments of this specification.

[0103] An embodiment of this specification further provides Figure 6 the schematic structural diagram of the electronic device shown. As Figure 6 , at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above method.

[0104] The systems, devices, modules or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0105] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, this specification can take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Moreover, this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0106] This specification is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0107] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device that implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0108] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0109] It should also be noted that the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising said element.

[0110] This specification may be described in the general context of computer-executable instructions executable by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media including storage devices.

[0111] Each embodiment in this specification is described in a progressive manner. For parts that are the same or similar among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, they are described relatively simply, and reference can be made to the corresponding parts of the method embodiments for related content.

[0112] The above description is only for the embodiments of this specification and is not intended to limit this specification. For those skilled in the art, various changes and modifications can be made to this specification. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this specification shall be included within the scope of the claims of this specification.

Claims

1. A method for training a result prediction model, comprising: Obtaining first sample training data in the timing scenario of outbound calls, wherein the first sample training data includes feature data of sample objects, a first decision item, and the factual result of the sample object regarding the first decision item. The sample object is a user, the feature data includes customer feature data collected at each time point for outbound calls, the first decision item includes the observed time point of making a call to the user, and the factual result includes the observed call answering result when making a call to the user at the time point represented by the first decision item; Training a fusion model based on the first sample training data to obtain a trained fusion model, wherein the fusion model includes a regression model and a propensity score model; Inputting a second decision item and the feature data into the trained fusion model to obtain a counterfactual result regarding the second decision item output by the regression model and a propensity score result output by the propensity score model; Constructing an unbiased augmented signal according to the first sample training data and the output of the fusion model for each decision item; Constructing second sample training data, wherein the second sample training data includes a third decision item and the feature data, and the third decision item includes the first decision item and the second decision item; Training a result prediction model according to the second sample training data and the unbiased augmented signal to obtain a trained result prediction model. Inputting the feature data corresponding to a target object and a target decision item into the trained result prediction model to obtain a prediction result corresponding to the target object, wherein the target object includes a user to be outbound called, the target decision item is used to indicate the time node of making a call, and the prediction result is used to predict whether the user to be outbound called will answer the call at this time node; Wherein, the constructing of the unbiased augmented signal includes: Constructing an augmented signal as follows: where \(t\) represents the \(t\)-th decision item, \(x\) represents the feature data of the sample object, and \(T\) represents the input decision item. represents the augmented signal constructed based on the estimated value of the regression model and the estimated value of the propensity model. represents the expected result of the regression model for the \(t\)-th decision item. represents the propensity score result of the propensity score model for the \(t\)-th decision item. The function takes the value of 1 when \(T = t\) and 0 when \(T\neq t\), and \(Y\) represents the factual result corresponding to the \(t\)-th decision item. Deriving an unbiased augmented signal by performing unbiasedness derivation on the augmented signal.

2. The method according to claim 1, wherein the training the fusion model based on the first sample training data to obtain a trained fusion model includes: Inputting the first decision item into the regression model in the fusion model to obtain an expected result corresponding to the first decision item output by the regression model, and constructing a first loss function based on the expected result and the observed result; Inputting the feature data into the propensity score model in the fusion model to obtain a propensity score result corresponding to the sample object output by the propensity score model, and constructing a second loss function based on the propensity score result and the first decision item; Training the fusion model according to the first loss function and the second loss function to obtain a trained fusion model.

3. The method according to claim 2, wherein the training the fusion model according to the first loss function and the second loss function to obtain a trained fusion model includes: Construct a third loss function based on the first loss function, the second loss function, and the balance parameter, and train the fusion model according to the third loss function; Adjust the balance parameter to converge the fusion model to obtain a trained fusion model.

4. The method according to claim 2, wherein the propensity score model comprises a first embedding layer and a first linear layer, the regression model comprises a second embedding layer and a second linear layer, and the input of the second linear layer comprises a second vector output by the second embedding layer and a first vector output by the first embedding layer.

5. The method according to claim 1, wherein training the result prediction model according to the second sample training data and the unbiased augmentation signal to obtain a trained result prediction model comprises: Inputting the second sample training data into the result prediction model to obtain a target expected result of the sample object with respect to the third decision item output by the result prediction model, and training the result prediction model by a fourth loss function constructed based on the target expected result and the unbiased augmentation signal to obtain a trained result prediction model.

6. The method according to claim 5, wherein the result prediction model comprises a third embedding layer, a fourth embedding layer, and a third linear layer, the input of the third embedding layer is the feature data, the input of the fourth embedding layer is the third decision item, and the input of the third linear layer comprises a third vector output by the third embedding layer and a fourth vector output by the fourth embedding layer.

7. A method for result prediction, comprising: In an outbound timing scenario, obtain feature data corresponding to a target object and a target decision item, wherein the target object comprises a user to be outbound called, and the target decision item is used to indicate a time node for making a call; Input the feature data and the target decision item into a trained result prediction model to obtain a prediction result corresponding to the target object, wherein the result prediction model is trained according to the method of any one of claims 1 to 6, and the prediction result is used to predict whether the user to be outbound called will answer the call at this time node.

8. A device for training a result prediction model, comprising: A first obtaining module, configured to obtain first sample training data in an outbound timing scenario, wherein the first sample training data comprises feature data of a sample object, a first decision item, and a factual result of the sample object with respect to the first decision item, the sample object is a user, the feature data comprises customer feature data collected at each time point for outbound calling, the first decision item comprises an observed time point for making a call to the user, and the factual result comprises an observed answer result for making a call to the user at the time point represented by the first decision item; A first training module, configured to train a fusion model based on the first sample training data to obtain a trained fusion model, wherein the fusion model comprises a regression model and a propensity score model; A second obtaining module, configured to input the second decision item and the feature data into the trained fusion model, and obtain a counterfactual result of the regression model for the second decision item and a propensity score result output by the propensity score model, where the second decision item is different from the first decision item; A first construction module, configured to construct an unbiased augmented signal according to the first sample training data and the output of the obtained fusion model for each decision item; A second construction module, configured to construct second sample training data, where the second sample training data includes a third decision item and the feature data, and the third decision item includes the first decision item and the second decision item; A second training module, configured to train a result prediction model according to the second sample training data and the unbiased augmented signal to obtain a trained result prediction model, where the feature data corresponding to the target object and the target decision item are input into the trained result prediction model to obtain a prediction result corresponding to the target object, where the target object includes a user to be called, the target decision item is used to indicate the time node of making a call, and the prediction result is used to predict whether the user to be called answers the call at this time node; Wherein, the constructing the unbiased augmented signal includes: Constructing an augmented signal as follows: where \(t\) represents the \(t\)-th decision item, \(x\) represents the feature data of the sample object, and \(T\) represents the input decision item. represents the augmented signal constructed based on the estimated value of the regression model and the estimated value of the propensity model. represents the expected result of the regression model with respect to the \(t\)-th decision item. represents the propensity score result of the propensity score model with respect to the \(t\)-th decision item. The function takes the value of 1 when \(T = t\) and 0 when \(T\neq t\), and \(Y\) represents the factual result corresponding to the \(t\)-th decision item. Deriving an unbiased augmented signal by performing unbiased derivation on the augmented signal.

9. An apparatus for result prediction, comprising: A third obtaining module, configured to obtain the feature data corresponding to the target object and the target decision item in the timing scenario of outbound calls, where the target object includes a user to be called, and the target decision item is used to indicate the time node of making a call; A fourth obtaining module, configured to input the feature data and the target decision item into a trained result prediction model to obtain a prediction result corresponding to the target object, where the result prediction model is trained based on the method according to any one of claims 1 to 7, and the prediction result is used to predict whether the user to be called answers the call at this time node.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

11. An electronic device, characterized in that, Including: A processor and a memory; wherein, the memory stores a computer program, and the computer program is adapted to be loaded and executed by the processor to implement the steps of the method according to any one of claims 1 to 7.

12. A computer program product having at least one instruction stored thereon, characterized in that, When the at least one instruction is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images