CN118014167A

CN118014167A - Method, apparatus, device, medium, and article for transformation prediction

Info

Publication number: CN118014167A
Application number: CN202410417898.6A
Authority: CN
Inventors: 於喆; 夏驰; 曹绍升; 周霖
Original assignee: Beijing Didi Infinity Technology and Development Co Ltd
Current assignee: Beijing Didi Infinity Technology and Development Co Ltd
Priority date: 2024-04-08
Filing date: 2024-04-08
Publication date: 2024-05-10

Abstract

In accordance with embodiments of the present disclosure, methods, apparatus, devices, media, and products for transformation prediction are provided. The method includes obtaining feature information associated with a target object in a target service demand to be processed and a candidate incentive scheme indicating incentive resources provided for the target object when processing the target service demand; determining, using the trained first machine learning model, a base conversion probability for the target object to perform a conversion behavior in the target service requirement without applying the candidate excitation scheme based on the feature information; determining, based on the feature information, a conversion probability boost value for the target object to perform a conversion action in the target service requirement with the candidate excitation scheme applied, using the trained second machine learning model; and determining a conversion probability of the target object performing the conversion behavior in the target service requirement in case of applying the candidate incentive scheme based on the base conversion probability and the conversion probability increase value.

Description

Method, apparatus, device, medium, and article for transformation prediction

Technical Field

Example embodiments of the present disclosure relate generally to the field of computers and, more particularly, relate to a method, apparatus, device, computer-readable storage medium and computer program product for transformation prediction.

Background

In many application scenarios, there are various service requirements to be handled, and there may be unbalance between supply and demand at different times and areas. For example, in the shared travel field, the service demand refers to the travel demand of passengers and the service of the travel demand by drivers. The unbalance of the supply and demand of the shared travel directly influences the travel market efficiency. The platform can accommodate market supply and demand imbalances by incentive strategies, but also needs to take into account cost factors. For a shared travel platform, it is desirable to adjust the supply-demand balance between the passengers of the vehicle and the driver servicing the passengers.

Disclosure of Invention

In a first aspect of the present disclosure, a method for transformation prediction is provided. The method comprises the following steps: obtaining feature information associated with a target object in a target service demand to be processed and a candidate incentive scheme indicating incentive resources provided for the target object when the target service demand is processed; determining, using the trained first machine learning model, a base conversion probability for the target object to perform a conversion behavior in the target service requirement without applying the candidate excitation scheme based on the feature information; determining, based on the feature information, a conversion probability boost value for the target object to perform a conversion action in the target service requirement with the candidate excitation scheme applied, using the trained second machine learning model; and determining a conversion probability of the target object performing the conversion behavior in the target service requirement in case of applying the candidate incentive scheme based on the base conversion probability and the conversion probability increase value.

In a second aspect of the present disclosure, an apparatus for transformation prediction is provided. The device comprises: an obtaining module configured to obtain feature information associated with a target object in a target service demand to be processed and a candidate incentive scheme indicating incentive resources provided for the target object when processing the target service demand; a base prediction module configured to determine a base conversion probability of the target object performing a conversion action in the target service requirement without applying the candidate excitation scheme based on the feature information using a trained first machine learning model; a boost prediction module configured to determine, based on the feature information, a conversion probability boost value for the target object to perform the conversion behavior in the target service requirement with the candidate excitation scenario applied, using a trained second machine learning model; and a transformation prediction module configured to determine a transformation probability that the target object performs the transformation behavior in the target service requirement if the candidate incentive scheme is applied based on the base transformation probability and the transformation probability elevation value.

In a third aspect of the present disclosure, an electronic device is provided. The apparatus comprises at least one processing unit; and at least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit. The instructions, when executed by at least one processing unit, cause the apparatus to perform the method of the first aspect.

In a fourth aspect of the present disclosure, a computer-readable storage medium is provided. The computer readable storage medium has stored thereon a computer program executable by a processor to implement the method of the first aspect.

In a fifth aspect of the present disclosure, a computer program product is provided. The computer program product comprises computer executable instructions which, when executed by a processor, implement the method of the first aspect.

It should be understood that what is described in this summary is not intended to limit the critical or essential features of the embodiments of the disclosure nor to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The above and other features, advantages and aspects of embodiments of the present disclosure will become more apparent by reference to the following detailed description when taken in conjunction with the accompanying drawings. In the drawings, wherein like or similar reference numerals denote like or similar elements, in which:

FIG. 1 illustrates a schematic diagram of an example environment in which embodiments of the present disclosure may be implemented;

FIG. 2 illustrates a schematic block diagram of a model framework for transformation prediction, according to some embodiments of the present disclosure;

FIG. 3 illustrates a schematic diagram of an environment for model training and application, according to some embodiments of the present disclosure;

FIG. 4 illustrates a flow chart of a process for conversion prediction according to some embodiments of the present disclosure;

FIG. 5 illustrates a block diagram of an apparatus for translation prediction according to some embodiments of the present disclosure; and

Fig. 6 illustrates a block diagram of an electronic device in which one or more embodiments of the disclosure can be implemented.

Detailed Description

Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While certain embodiments of the present disclosure have been illustrated in the accompanying drawings, it is to be understood that the present disclosure may be embodied in various forms and should not be construed as limited to the embodiments set forth herein, but rather, these embodiments are provided so that this disclosure will be more thorough and complete. It should be understood that the drawings and embodiments of the present disclosure are for illustration purposes only and are not intended to limit the scope of the present disclosure.

In describing embodiments of the present disclosure, the term "comprising" and its like should be taken to be open-ended, i.e., including, but not limited to. The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". Other explicit and implicit definitions are also possible below. The terms "first," "second," and the like, may refer to different or the same object. Other explicit and implicit definitions are also possible below.

Embodiments of the present disclosure may relate to user data, the acquisition and/or use of data, and the like. These aspects all follow corresponding legal and related regulations. In embodiments of the present disclosure, all data collection, acquisition, processing, forwarding, use, etc. is performed with knowledge and confirmation by the user. Accordingly, in implementing the embodiments of the present disclosure, the user should be informed of the type of data or information, the range of use, the use scenario, etc. that may be involved and obtain the authorization of the user in an appropriate manner according to the relevant laws and regulations. The particular manner of notification and/or authorization may vary depending on the actual situation and application scenario, and the scope of the present disclosure is not limited in this respect.

The term "responsive to" as used herein means a state in which a corresponding event occurs or a condition is satisfied. It will be appreciated that the execution timing of a subsequent action that is executed in response to the event or condition is not necessarily strongly correlated with the time at which the event occurs or the condition is established. For example, in some cases, the follow-up actions may be performed immediately upon occurrence of an event or establishment of a condition; in other cases, the subsequent action may be performed after a period of time has elapsed after the event occurred or the condition was established.

As used herein, the term "model" may learn the association between the respective inputs and outputs from training data so that, for a given input, a corresponding output may be generated after training is completed. The generation of the model may be based on machine learning techniques. Deep learning is a machine learning algorithm that processes inputs and provides corresponding outputs through the use of multiple layers of processing units. The "model" may also be referred to herein as a "machine learning model," "machine learning network," or "network," and these terms are used interchangeably herein.

Generally, machine learning may generally include three phases, namely a training phase, a testing phase, and an application phase (also referred to as an inference phase). In the training phase, a given model may be trained using a large amount of training data, iteratively updating parameter values until the model is able to obtain consistent inferences from the training data that meet the desired goal. By training, the model may be considered to be able to learn the association between input and output (also referred to as input to output mapping) from the training data. Parameter values of the trained model are determined. In the test phase, test inputs are applied to the trained model to test whether the model is capable of providing the correct output to determine the performance of the model. The test phase may sometimes be fused in the training phase. In the application or reasoning stage, the trained model may be used to process the actual model input based on the trained parameter values to determine the corresponding model output.

As mentioned previously, in different service demand scenarios, to accommodate supply and demand imbalance, supply and demand imbalance issues may be accommodated by an incentive strategy (also referred to as a subsidy strategy). Incentive policies refer to providing incentive resources to an object to which a service demand relates, desirably to be able to incentive the object to perform a transformation action on the service demand. For example, for a demander of a service demand, the transformation behavior refers to initiating the service demand; for the recipient of the service demand, the transformation action refers to providing the required service demand. Taking a shared travel service involving a passenger and a driver as an example, the conversion behavior refers to the passenger initiating a travel order and the driver accepting the travel order initiated by the passenger. Incentive resources may include subsidies to the passenger and/or to the driver's order.

Cost factors are also typically considered when providing incentive strategies. Thus, causal inference can be employed to determine how to provide the appropriate incentive resources and obtain the expected probability of conversion.

FIG. 1 illustrates a schematic diagram of an example environment 100 in which embodiments of the present disclosure may be implemented. The application of incentive policies in a shared travel scenario is explained in the environment 100 by taking presence services as an example.

On the passenger side, the passenger 112 enters (e.g., by way of in a corresponding application) a consultation of travel needs. Travel requirements include the start and end of travel, and possibly other relevant information (e.g., travel time, travel preferences, etc.). Travel demands are sometimes also referred to as travel orders or ride orders.

On the platform side, a demand processing platform 120 (which may correspond to a platform providing shared travel services, sometimes also referred to simply as a "platform") generates incentive schemes for passengers 112 from passenger incentive engines 122 based on travel demand. An incentive scheme for the passenger 112 may include a combination of multiple incentive resources provided to the passenger when one or more travel patterns are utilized to handle the current travel demand. Different travel patterns may correspond to different types of shared vehicles in the shared travel service. Incentive resources may include, for example, price subsidies, virtual resource subsidies, and the like.

The passenger 112 may decide whether to place an order, i.e., issue a downstream demand to the demand processing platform 120, based on the current incentive scheme. In the event that the passenger 112 issues a travel demand to the demand processing platform 120, the demand processing platform 120 adds the travel demand to the order list until a location is matched with the appropriate driver.

The demand processing platform 120 may also determine the idle vehicles 132 based on the positioning information and use a particular matching strategy to match each driver to the demand of the passengers 112. In some cases, the demand processing platform 120 may generate an incentive scheme for the driver from the driver incentive engine 124 based on current travel demand. The incentive scheme for the driver may include a combination of a plurality of incentive resources provided to the driver when one or more travel patterns are utilized to handle the current travel demand. Incentive resources may include, for example, price subsidies, virtual resource subsidies, and the like.

On the driver side, the driver can determine whether the travel demands of the passengers are to be met according to the given incentive scheme. When the driver decides to accept the current travel demand, the driver will be removed from the list of free drivers and will reach the start of the journey specified by the passenger in the travel demand. The passenger and driver will begin the journey.

Under the scene of sharing travel, causal inference between motivation resources and conversion behaviors, namely reasoning of the conversion behaviors of passengers and drivers under different motivation schemes, plays an important role in the scene of adjusting the requirement of sharing travel. The causal inference model may be trained using machine learning techniques to perform this task.

In the scheme of estimating the conversion behavior estimation of the platform passenger and the driver under different incentives, the following properties need to be considered. First, the non-uniformity and correlation of the stimulus distribution among users is considered. Historically, motivational resource allocation has often been correlated to user characteristics, such as a tendency to inactive users, etc., which has resulted in bias in direct estimation results. In addition, order context characteristics need to be considered. This is because each travel order has different characteristics that affect the sensitivity of the type of order to incentive resources, and the incentive resources are sent to the most elastic order to maximize the order volume of the lift platform. Third, the excitation boost effect is also considered. For resource incentives, it is necessary to know not only the reaction of user behavior under different resource incentives, but also the promotion of user transformation behavior after incentives. For example, when giving incentive resources to drivers, it is equally important which drivers are on-line and which drivers are more up through incentive.

Causal inference modeling is currently performed mainly by the following schemes. In the first approach, causal inference models are trained by directly using historical data, without regard to the correlation between the excitation profile and the sample features. In the first scenario, data is collected by random trial. Excitation resources are first randomly issued and modeled directly using random experimental data. In a third approach, historical data is used to train a causal inference model while estimating whether a sample will get excited and the transformation behavior of the sample under excitation.

The causal inference modeling described above suffers from shortcomings in different respects, principally in the following respects. Modeling directly using historical data can cause the model to deviate significantly from the correct estimate. The random experiment cost is higher, the experiment time is longer, and the obtained effective data is limited. Furthermore, both of these approaches have difficulty in both predicting behavior and predicting behavior improvement, and in dealing with continuous incentives.

Embodiments of the present disclosure propose a scheme for conversion prediction in a demand processing scenario. According to various embodiments of the present disclosure, in a model, a transformation behavior prediction is divided into two parts, one part being the probability that a target object performs a transformation behavior when not excited; another part characterizes the probability improvement of the transformation behavior due to the excitation. The model framework can correct the correlation between the excitation and the characteristic information of the object, and can also consider the prediction precision of the transformation behavior and the prediction precision of the transformation probability improvement. In some embodiments, the model is designed by the loss function of the model such that the predictions in the two aspects are as independent (orthogonal) of each other as possible in training. In this way, where the causal inference model employed by embodiments of the present disclosure is applicable to the uneven distribution of stimulus distribution in historical data used for training, an accurate estimate of the improvement in transformation behavior is made.

Hereinafter, for ease of discussion, causal inference between incentive resource issuance and conversion behavior prediction is described by taking a shared travel service as an example. However, it should be appreciated that embodiments of the present disclosure may be applied to any other service-requiring application scenario where it is desirable to predict whether various types of objects will perform the intended transformation behavior under different incentive schemes.

Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings.

FIG. 2 illustrates a schematic block diagram of a model framework 200 for transformation prediction according to some embodiments of the present disclosure. The model framework 200 of fig. 2 is a causal inference model based on a multi-tasking deep neural network.

The model framework 200 includes three portions, a first machine learning model 210 (denoted asModel), a second machine learning model 220, and a reference machine learning model 226. The first machine learning model 210 is configured to determine a base transformation probability of the target object performing a transformation behavior in the target service requirement without applying the candidate excitation scheme (i.e., t=0). The first machine learning model 210 may also sometimes be referred to as a base conversion prediction model. The second machine learning model 220 is configured to determine a conversion probability increase value for the target object to perform a conversion action in the target service requirement if the candidate incentive scheme is applied, and thereby determine a conversion probability/>, for the target object to perform a conversion action in the target service requirement if the candidate incentive scheme is applied. The second machine learning model 220 may also sometimes be referred to as a conversion probability prediction model.

As will be mentioned below with respect to the description of model training, the reference machine learning model 226 primarily participates in the training of the first machine learning model 210 and the second machine learning model 220. After model training, causal inference of incentive resource issuance and transformation behavior prediction may be achieved using the trained first machine learning model 210 and second machine learning model 220.

The following describes the use of the trained first machine learning model 210 and second machine learning model 220 to infer a transformation probability of a target object performing a transformation behavior in a target demand with the application of a certain incentive scheme.

The mode input for the entire module framework 220 includes feature information X associated with the target object in the target service demand to be processed and the candidate incentive scheme.

A target object refers to an object involved in a target service requirement, including an initiator or provider of the target service requirement. The characteristic information X associated with the target object may comprise any suitable relevant information related to the target object and/or the target demand. The selection of the feature information X in the embodiments of the present disclosure is not particularly limited.

Taking an application scenario of sharing travel as an example, the target service requirement may include a travel requirement, and the target object may include a passenger initiating the travel requirement or a driver accepting the travel requirement. In some embodiments, the first machine learning model 210 and the second machine learning model 220 may be trained separately for the passenger side and the driver side, respectively, i.e., with the model framework 200, respectively. The first machine learning model 210 and the second machine learning model 220 trained for the passenger side may be applied in the passenger incentive engine 122 of fig. 1 for assisting in determining an incentive scheme for the passenger. The first machine learning model 210 and the second machine learning model 220 trained for the driver side may be applied in the driver incentive engine 122 of FIG. 2 for assisting in determining an incentive scheme for the driver.

In some embodiments, for travel demand, the characteristic information X associated with the passenger includes, but is not limited to, starting point information of the travel demand, ending point information of the travel demand, vehicle supply information of the starting point of the travel demand, vehicle supply information of the ending point of the travel demand, historical demand information of the passenger (e.g., historical vehicle order initiation behavior of the passenger), and so forth. The characteristic information X associated with the driver includes, but is not limited to, start point information of travel demand, end point information of travel demand, vehicle supply information of start point of travel demand, vehicle supply information of end point of travel demand, history demand information of the driver (e.g., history of vehicle order service behavior of the driver), and the like. It should be understood that only a few examples are given here, and that more, fewer, or other types of characteristic information may be devised in practical applications.

The candidate incentive scheme indicates incentive resources T provided for the target object when processing the target service demand. One or more processing categories may be utilized in processing the target service demand to process the target service demand. When there are multiple processing categories, the candidate incentive scheme may include a combination of multiple incentive resources that are respectively provided for the target object when the target service demand is processed with the multiple processing categories. For example, in a travel scenario, the candidate incentive scheme may include a combination of incentive resources provided for the passenger or driver, respectively, when processing travel demands using one or more travel patterns. Different travel patterns may include different types of vehicles. This is to take into account that under different processing categories (e.g. different modes of travel) different incentive resources may be provided, as well as different probabilities of the target object performing the transformation behaviour under a combination of these incentive resources. Candidate incentive schemes that may be considered may include the total combined number of incentive resources under different processing categories.

Here, the target object's execution of the transformation behavior for the target demand has a relationship with the target object's role in the demand supply. For example, for a passenger, the conversion behavior for travel demand includes placing a travel order for travel demand; for the driver, the translating action for travel demand includes accepting a travel order initiated by the passenger.

In the example of FIG. 2, for purposes of explanation, it is assumed that there are two processing categories, the excitation resources corresponding to the two processing categories are represented as,/>. In some embodiments,/>,/>May be a continuous variable, and may be selected from a range of continuous excitation resource values corresponding to each of the two processing categories. /(I)Indicating that no incentive resource is applied. In other embodiments, the model architecture 200 may have one incentive resource corresponding to one processing class, or may have more than two incentive resources corresponding to more than one processing class, as embodiments of the present disclosure do not limit this.

In performing transformation behavior prediction, it may be desirable to predict the probability that a target object will perform transformation behavior under different incentive schemes. In this case, the trained first machine learning model 210 and second machine learning model 220 may be utilized multiple times to predict the probability of transformation of the target object under multiple candidate excitation scenarios. Based on the probability of transformation under the different candidate excitation schemes, the target excitation scheme issued to the target object may be finally determined. Of course, the embodiments of the present disclosure do not impose specific restrictions on the selection of subsequent excitation schemes, etc., but rather how the primary relationship predicts the probability of transformation of the target object given the excitation scheme.

In the model process, the feature information X may first be input into the feature network 202 separately to extract an embedded representation of the feature information X (embedding). In some embodiments, the feature network 220 may include a multi-layer perceptron MLP, although the feature information X may be converted into a form of embedded representation that is understandable to the model based on other network structures.

The embedded representation of the feature information X is input into the trained first machine learning model 210 and the second machine learning model 220. The trained first machine learning model 210 determines a base transformation probability for the target object to perform a transformation action in the target service requirement without applying the candidate incentive scheme based on the characteristic information X (more specifically, an embedded representation of the characteristic information X). The first machine learning model 210 may include a feature extraction portion for extracting a base conversion probability b 212 from the embedded representation of the input feature information X, and then the first machine learning model 210 may map the base conversion probability b 212 to a base conversion probability by a mapping function, such as a sigmoid layer 214. The sigmoid layer 214 may map the base conversion probability b 212 into a value range of 0 to 1 using a sigmoid function, expressed as. It should be appreciated that the base conversion probability b 212 and the specific base conversion probability/>Is basically equivalent, but the value ranges are different.

Thus, b represents the natural behavioral component of the target object predicted without stimulus, after passing through the sigmoid layer 214,Is a prediction of the transformation behavior without excitation mapped into a range of probability values.

In some embodiments, the feature extraction portion of the first machine learning model 210 and/or the second machine learning model 220 mentioned later may include a multi-layer perceptron MLP, although the feature information X may be converted into a form of embedded representation that is understandable to the model based on other network structures.

The trained second machine learning model 220 determines a conversion probability boost value for the target object to perform a conversion action in the target service requirement if the candidate incentive scheme is applied based on the feature information X (more specifically, an embedded representation of the feature information X).

In some embodiments where the incentive scheme includes a combination of multiple incentive resources that are respectively provided for the target object when the target service demand is processed with multiple processing categories, the second machine learning model 220 includes multiple sub-models that respectively correspond to the multiple processing categories. With each of the plurality of sub-models, a sub-conversion probability elevation value for the target object to perform a conversion behavior in the target service requirement with the application of the incentive resource of the corresponding processing class is determined based on the characteristic information X (more specifically, an embedded representation of the characteristic information X). In the example of fig. 2, the second machine learning model 220 includes and incentives resourcesA corresponding feature extraction section for determining excitation resources/>, from the embedded representation of the input feature information XCorresponding weight value/>224-1, And with incentive resources/>A corresponding feature extraction section for determining excitation resources/>, from the embedded representation of the input feature information XCorresponding weight value/>224-2. Of course, if more processing categories exist, the second machine learning model 220 may also continue to be extended in a similar manner.

In some embodiments, in the second machine learning model 220, after extracting the intermediate feature representation from the embedded representation of the feature information X, maps the intermediate feature representation to a weight value, e.g., a weight value, using a mapping function, e.g., mapping layers 222-1, 222-2And/>. The weight values are used to weight the corresponding incentive resources (which may be represented as incentive values), e.g./>The conversion probability under the corresponding excitation resource is improved.

In some embodiments, feature extraction portions corresponding to different excitation resources may each extract a respective intermediate feature representation for determining the weight value. The excitation resources are always integer values and the application of the excitation resources will also have a positive effect on the conversion probability, so the mapping functions 222-1, 222-2 are configured to map the intermediate feature representation to positive weight values. For example, mapping to 222-1, 222-2 may be performed by a softplus function such that the weight valuesAnd/>Always positive.

The second machine learning model 220 determines a transition probability boost value in the case of applying the candidate excitation scheme based on the base transition probability and a plurality of sub-transition probability boost values corresponding to the plurality of processing categories. Specifically, the second machine learning model 220 may be composed of the aggregate underlying transformation probabilities b 212224-1 And/>224-2, And mapping the aggregated embedded identity into a value range of 0 to 1 by means of a mapping function, e.g. sigmoid layer 226, to obtain a transformation probability/>, of the target object performing a transformation action in the target service requirement if the candidate incentive scheme is applied. In some embodiments, the probability of conversion/>The determination of (2) may be expressed as follows:

In the above-mentioned formula (1), In the absence of application of candidate excitation schemes (/ >)) Basic transformation probability of target object in case of (a)/>Representing the transition probability boost value of the target object in case of applying the candidate scheme,/>Is represented in the application of candidate excitation schemes (/ >)) Transformation probability of target object in case of (a) and/>Representing a conversion probability (e.g., a probability that a user initiates a travel order or a driver accepts a travel order) mapped to a range of values from 0 to 1. Thus,/>Representation prediction in applying incentive resources/>In case of target object conversion probability improvement condition,/>Representation prediction in applying incentive resources/>The transformation probability of the target object increases the situation. In some embodiments, the total conversion probability/>Conversion probability b with basis, incentive resource/>Corresponding conversion probability promotion and incentive resource/>The corresponding transition probability increases are linear.

The model framework shown in fig. 2 is suitable for causal inference without random experimental data. Based on the model framework of fig. 2, two objectives can be met by training, namely taking into account the relation of the stimulus and the characteristic information present in the data, and also taking into account the predictions of the transformation behavior and the predictions of the transformation probability improvement.

How to train the first machine learning model 210 and the second machine learning model 220 will be described further below.

The training samples used for training may be historical data collected from historical demand processing. For a particular service requirement performed by a particular object in the historical data, each training sample may include sample characteristic information for sample object iConversion tag of sample object/>. During the history service demand process, a single sample object i may or may not be provided with some incentive scheme, and may or may not perform a transformation action (e.g., a travel order may not be placed, or may not be accepted). Depending on the actual situation of sample object i, the transformation tag/>Either indicating the actual conversion situation where no incentive scheme is applied when handling the service requirements of the sample object i or indicating the actual conversion situation where a certain incentive scheme is applied. The real excitation scheme corresponding to sample object i is denoted/>Wherein/>Meaning that sample object i is not provided with any excitation scheme,/>Meaning that the sample object i is provided with a given excitation scheme. Conversion tag/>The specific value of (2) may be/>=0, Indicating that sample object i actually does not perform the transformation behavior, or/>=1, Indicating that sample object i actually performs the transformation behavior. The number of training samples used for training may be high to cover situations where conversion behavior occurs or does not occur, as well as to cover various possible incentive schemes, such as various combinations of incentive resources under multiple processing categories.

In some embodiments, the overall loss function of the first machine learning model 210 and the second machine learning model 220 may be defined as follows:

In the above equation (2), the first loss function Is configured to measure a conversion probability prediction loss of the first machine learning model 210 and the second machine learning model 220. Assuming that the total number of training samples is n,/>Can be defined as follows:

Either the predicted base conversion probability determined using the first machine learning model 210 being trained or the predicted conversion probability determined using the first machine learning model 210 being trained and the second machine learning model 220, depending on whether the sample object i is applied with a certain incentive strategy, i.e., incentive strategy/> Or/>。

For the sample object i corresponding to each training sample, a first loss functionBased on the loss of: in the event that the conversion label indicates a true conversion condition for the non-applied incentive scheme, the predicted base conversion probability determined using the first machine learning model 210 being trained and the true conversion condition/>, for the non-applied incentive schemeError between, or in the case where the conversion label indicates a true conversion condition for which the excitation scheme is applied, the predicted conversion probability determined using the first machine learning model 210 and the second machine learning model 220 being trained and the true conversion condition/>, for which the excitation scheme is appliedErrors between them. First loss function/>May be expressed as a predicted base conversion probability or a predicted conversion probability/>And true transformation status/>Cross entropy loss between.

In the above equation (2), the second loss functionIs configured to measure the prediction error of the excitation pattern T for the object. The penalty of the second penalty function is constructed as the error between the predicted excitation pattern application determined based on the sample characteristic information using the reference machine learning model 230 being trained and the actual excitation pattern application of the sample object. Second loss functionCan be defined as follows:

As shown in fig. 2, the input of the reference machine learning model 230 is feature information (or embedded representation corresponding to the feature information) of the object, and the reference machine learning model 230 outputs predictions of the excitation scheme for the object through the feature extraction section and the output sigmoid layer 232 . For example, the output of the reference machine learning model 230 may include probabilities of applying all possible incentive schemes to the input object, and the incentive scheme with the highest probability may be confirmed to be the recommended incentive scheme for the input object. The reference machine learning model 230 may also sometimes be referred to as an excitation prediction model. Second loss function/>Predicted excitation scheme/>, constructed as output for sample object i with reference to machine learning model 230Real excitation scheme/>, corresponding to sample object iCross entropy loss between.

In addition to the above first and second loss functions, the training of the first machine learning model 210 and the second machine learning model 220 is also based on a third loss functionThe/>Can be considered as an independence regularization term for making/>Is statistically independent of the prediction error of T. When training samples are constructed by using historical data, the conversion behavior of the object without excitation and the conversion probability improvement caused by the excitation cannot be simultaneously acquired aiming at the same object and the same service requirement. The introduction of the independence regularization term in the training process can enable the prediction of the conversion probability and the prediction of the excitation scheme to be independent of each other, so as to overcome estimation deviation caused by uneven excitation distribution in training data.

Specifically, a third loss functionCan be defined as follows:

For the sample object i corresponding to each training sample, a third loss function Based on the loss of: prediction error of the conversion probability in case that the excitation scheme is not applied to the sample object, and prediction error applied to the excitation scheme of the sample object. Specifically, when the above formula (8) is brought into the above formula (5) (/ >)) A prediction error representing a conversion probability in the case where the sample object is not applied with the excitation scheme,For measuring the prediction error applied to the excitation scheme for the sample object. Prediction error for excitation scheme application/>Application of/>, based on predictive excitation scheme determined based on sample feature information using the training reference machine learning model 230True excitation scheme application with sample object/>Errors between them.

Representing the predicted transition probability in case the sample object i is not applied with any excitation scheme,Calculated by the above formula (6)/>. In case sample object i is not actually stimulated by the application,/>=0, Then/>. That is, in the case where the conversion tag indicates a real conversion condition in which the excitation scheme is not applied to the sample object i,/>The value of (2) is the true conversion status of the sample object i. At this time, (/ >)) Representing predicted underlying transformation probabilities/>, determined using the first machine learning model 210 being trainedTrue transformation status/>, with no excitation scheme appliedErrors between them. In case the sample object i is actually stimulated by application,/>=1, Then/>Equal to the estimated base conversion probability in case it is assumed that sample object i has not taken a patch, at this time (/ >)) Representing predicted underlying conversion probability/>, determined using the first machine learning model 210 being trained, in the event that the conversion label indicates a true conversion condition for which the excitation scheme is appliedAnd an estimated base conversion probability/>, determined for sample object i, when no excitation scheme is appliedErrors between them. That is, estimating the underlying conversion probability is based on the true conversion condition/>, where the excitation scheme is appliedAnd the mean/>, of the transition probability rise values for all n training samples using the second machine learning model 220Differences between them. The equation (7) above gives the way the mean value of the transition probability rise values for n training samples is calculated.

Statistically independent or statistically uncorrelated between the two random numbers is in fact equal to zero, the product of the two. In the total loss functionIf pair/>The derivative is then changed to "multiply" the error of the transition probability without excitation by the prediction error at the excitation scheme. By minimizing the total loss function/>Can let pair/>The result of the derivative is equal to or close to 0, thereby realizing statistical independence or statistical uncorrelation of the two items.

During the model training process, the first machine learning model 210, the second machine learning model 220, and the reference machine learning model 230 may be co-trained and updated. In some embodiments, the reference machine learning model 230 may be discarded after training is complete. In some embodiments, the reference machine learning model 230 may also be continued to be used for prediction of the excitation scheme. Embodiments of the present disclosure are not limited in this regard.

FIG. 3 illustrates a schematic diagram of an environment 300 for model training and application, according to some embodiments of the present disclosure. In the environment 300 of fig. 3, the model is generally shown to involve different stages, including a training stage 302 and an application stage 306. There may also be a test phase after the training phase is completed, not shown in the figure.

In training phase 302, model training system 310 is configured to perform training of model 305 using training data set 332. Model 305 may be based on model framework 200 in fig. 2. At the beginning of training, the model may have initial parameter values. The training process is to update the parameter values of the model 305 to desired values based on the training data.

In the application phase 306, the obtained model 305 has trained parameter values that can be provided to the model application system 330 for use. In the application stage 306, the model 305 may be utilized to process corresponding target inputs 332 in the actual scene and provide corresponding target outputs 334. Model application system 330 may be configured to implement demand processing platform 120 of fig. 1. In such an embodiment, the target input 332 of the model application system 330 includes characteristic information of the target object in the target demand, as well as candidate excitation patterns to be predicted; the target output 334 includes a conversion probability that the target object performs a conversion action in the target service requirement if the candidate incentive scheme is applied. The model application system 330 may utilize the first machine learning model 210 and the second machine learning model 220 in the model framework 200 of fig. 2 to determine a target output based on the target input 332.

In fig. 3, model training system 310 and model application system 330 may include any computing system having computing capabilities, such as various computing devices/systems, terminal devices, servers, etc. The terminal device may relate to any type of mobile terminal, fixed terminal, or portable terminal, including mobile handsets, desktop computers, laptop computers, notebook computers, netbook computers, tablet computers, media computers, multimedia tablets, or any combination of the foregoing, including accessories and peripherals of these devices, or any combination thereof. Servers include, but are not limited to, mainframes, edge computing nodes, computing devices in cloud environments, and the like.

It should be appreciated that the components and arrangements in environment 300 shown in fig. 3 are merely examples, and that a computing system suitable for implementing the example implementations described in this disclosure may include one or more different components, other components, and/or different arrangements. For example, while shown as separate, model training system 310 and model application system 330 may be integrated in the same system or device. Implementations of the present disclosure are not limited in this respect.

FIG. 4 illustrates a flow chart of a process 400 for conversion prediction according to some embodiments of the present disclosure. Process 400 may be implemented at demand processing platform 120 of fig. 1.

At block 410, the demand processing platform 120 obtains feature information associated with a target object in the target service demand to be processed, as well as a candidate incentive scheme indicating incentive resources provided for the target object when processing the target service demand.

At block 420, the demand processing platform 120 determines a base conversion probability for the target object to perform a conversion behavior in the target service demand based on the feature information using the trained first machine learning model without applying the candidate incentive scheme.

At block 430, the demand processing platform 120 determines a conversion probability boost value for the target object to perform a conversion action in the target service demand with the candidate incentive scheme applied based on the feature information using the trained second machine learning model.

At block 440, the demand processing platform 120 determines a conversion probability that the target object performs a conversion action in the target service demand with the candidate incentive scheme applied based on the base conversion probability and the conversion probability boost value.

In some embodiments, the candidate incentive scheme includes a combination of multiple incentive resources that are respectively provided for the target object when utilizing multiple processing categories to process the target service demand.

In some embodiments, the second machine learning model includes a plurality of sub-models corresponding to a plurality of process categories, respectively. In some embodiments, determining a conversion probability boost value for a target object to perform a conversion action in a target service requirement if the candidate incentive scheme is applied comprises: determining, with each of the plurality of sub-models, a sub-conversion probability increase value for the target object to perform a conversion behavior in the target service requirement in the case of applying the incentive resource of the corresponding processing class based on the feature information; and determining a transition probability boost value in the case of applying the candidate excitation scheme based on the plurality of sub-transition probability boost values determined by the plurality of sub-models.

In some embodiments, determining a conversion probability boost value for a target object to perform a conversion action in a target service requirement if the candidate incentive scheme is applied comprises: extracting, from the feature information, intermediate feature representations in the case of applying the candidate excitation scheme, using a second machine learning model; mapping the intermediate feature representation to a weight value for the candidate excitation pattern using a mapping function in the second machine learning model; and determining a transition probability boost value based on the weight value and the excitation resources indicated by the candidate excitation scheme.

In some embodiments, the mapping function is configured to map the intermediate feature representation to a positive weight value.

In some embodiments, the first machine learning model and the second machine learning model are trained using a plurality of training samples including sample characteristic information of a sample object, a conversion label of the sample object indicating a true conversion condition in which an incentive scheme is not applied or a true conversion condition in which the incentive scheme is applied when processing service requirements of the sample object.

In some embodiments, the training of the first machine learning model and the second machine learning model is further based on the following loss functions: a first loss function, the loss of which is based on, for each training sample: the method may further include using an error between a predicted base conversion probability determined by the first machine learning model being trained and the actual conversion condition of the non-applied excitation scheme, if the conversion label indicates the actual conversion condition of the non-applied excitation scheme, or using an error between a predicted conversion probability determined by the first and second machine learning models being trained and the actual conversion condition of the applied excitation scheme, if the conversion label indicates the actual conversion condition of the applied excitation scheme. The training of the first machine learning model and the second machine learning model is further based on the following penalty functions: a second loss function, the loss of which is constructed as: an error between the predicted excitation pattern application determined based on the sample characteristic information and the actual excitation pattern application of the sample object using the reference machine learning model being trained.

In some embodiments, the training of the first machine learning model and the second machine learning model is based on the following loss functions: third loss function, for each training sample, the loss of the third loss function is based on: prediction error of the conversion probability in case that the excitation scheme is not applied to the sample object, and prediction error applied to the excitation scheme of the sample object.

In some embodiments, the prediction error of the conversion probability comprises: in the case where the conversion label indicates a true conversion condition for which the excitation scheme is not applied, the error between the predicted base conversion probability determined using the first machine learning model being trained and the true conversion condition for which the excitation scheme is not applied, or in the case where the conversion label indicates a true conversion condition for which the excitation scheme is applied, the error between the predicted base conversion probability determined using the first machine learning model being trained and the estimated base conversion probability determined for the sample object when the excitation scheme is not applied, the estimated base conversion probability being based on a difference between the true conversion condition for which the excitation scheme is applied and a mean of a plurality of conversion probability increase values for a plurality of training samples using the second machine learning model. In some embodiments, the prediction error applied by the excitation scheme includes: an error between the predicted excitation pattern application determined based on the sample characteristic information and the actual excitation pattern application of the sample object using the reference machine learning model being trained.

In some embodiments, the target service demand includes a travel demand, and the target object includes a passenger initiating the travel demand or a driver accepting the travel demand. In some embodiments, the first machine learning model and the second machine learning model are trained for the passenger side or for the driver side.

In some embodiments, the candidate incentive scheme includes a combination of incentive resources provided for the passenger or driver, respectively, when utilizing one or more travel patterns to handle travel demands.

In some embodiments, the characteristic information associated with the target object includes at least one of: the travel request includes start point information of the travel request, end point information of the travel request, vehicle supply information of the start point of the travel request, vehicle supply information of the end point of the travel request, history demand information of the passenger, or history demand information of the driver.

Fig. 5 illustrates a schematic block diagram of an apparatus 500 for translation prediction according to some embodiments of the present disclosure. The apparatus 500 may be implemented as or included in the demand processing platform 120 of fig. 1, the model training system 310 of fig. 3, and/or the model application system 330. The various modules/components in apparatus 700 may be implemented in hardware, software, firmware, or any combination thereof.

As shown in fig. 5, the apparatus 500 includes an obtaining module 510 configured to obtain feature information associated with a target object in a target service demand to be processed and a candidate incentive scheme indicating incentive resources provided for the target object when processing the target service demand. The apparatus 500 further comprises a base prediction module 520 configured to determine a base conversion probability of the target object performing the conversion behavior in the target service requirement without applying the candidate incentive scheme based on the feature information using the trained first machine learning model.

The apparatus 500 further comprises a boost prediction module 530 configured to determine a conversion probability boost value for the target object to perform a conversion behavior in the target service requirement with the candidate incentive scheme applied based on the feature information using the trained second machine learning model. The apparatus 500 further comprises a conversion prediction module 550 configured to determine a conversion probability of the target object performing the conversion behavior in the target service requirement in case the candidate incentive scheme is applied based on the base conversion probability and the conversion probability boost value.

In some embodiments, the second machine learning model includes a plurality of sub-models corresponding to a plurality of process categories, respectively. In some embodiments, the lift prediction module 530 includes: a sub-boost value determination module configured to determine, based on the feature information, a sub-conversion probability boost value for the target object to perform a conversion behavior in the target service requirement with application of the incentive resource of the corresponding processing class, using each of the plurality of sub-models; and a total boost value determination module configured to determine a transition probability boost value in the case of applying the candidate excitation scheme based on the plurality of sub-transition probability boost values determined by the plurality of sub-models.

In some embodiments, the lift prediction module 530 includes: a feature extraction module configured to extract, from the feature information, an intermediate feature representation in the case of applying the candidate excitation scheme, using the second machine learning model; a weight determination module configured to map the intermediate feature representation to weight values for the candidate excitation pattern using a mapping function in the second machine learning model; and a weight-based boost prediction module configured to determine a transition probability boost value based on the weight value and the excitation resources indicated by the candidate excitation scheme.

The elements and/or modules included in apparatus 500 may be implemented in various manners, including software, hardware, firmware, or any combination thereof. In some embodiments, one or more units and/or modules may be implemented using software and/or firmware, such as machine executable instructions stored on a storage medium. In addition to or in lieu of machine-executable instructions, some or all of the units and/or modules in apparatus 500 may be implemented at least in part by one or more hardware logic components. By way of example and not limitation, exemplary types of hardware logic components that can be used include Field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standards (ASSPs), systems On Chip (SOCs), complex Programmable Logic Devices (CPLDs), and the like.

Fig. 6 illustrates a block diagram that shows an electronic device 600 in which one or more embodiments of the disclosure may be implemented. It should be understood that the electronic device 600 illustrated in fig. 6 is merely exemplary and should not be construed as limiting the functionality and scope of the embodiments described herein. The electronic device 600 illustrated in fig. 6 may be used to implement the demand processing platform 120 of fig. 1, the model training system 310 and/or the model application system 330 of fig. 3, or the apparatus 400 of fig. 4.

As shown in fig. 6, the electronic device 600 is in the form of a general purpose computing device. The components of electronic device 600 may include, but are not limited to, one or more processors or processing units 610, memory 620, storage 630, one or more communication units 640, one or more input devices 650, and one or more output devices 660. The processing unit 610 may be an actual or virtual processor and is capable of performing various processes according to programs stored in the memory 620. In a multiprocessor system, multiple processing units execute computer-executable instructions in parallel to increase the parallel processing capabilities of electronic device 600.

The electronic device 600 typically includes a number of computer storage media. Such a medium may be any available media that is accessible by electronic device 600, including, but not limited to, volatile and non-volatile media, removable and non-removable media. The memory 620 may be volatile memory (e.g., registers, cache, random Access Memory (RAM)), non-volatile memory (e.g., read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory), or some combination thereof. Storage device 630 may be a removable or non-removable media and may include machine-readable media such as flash drives, magnetic disks, or any other media that may be capable of storing information and/or data (e.g., training data for training) and may be accessed within electronic device 600.

The electronic device 600 may further include additional removable/non-removable, volatile/nonvolatile storage media. Although not shown in fig. 6, a magnetic disk drive for reading from or writing to a removable, nonvolatile magnetic disk (e.g., a "floppy disk") and an optical disk drive for reading from or writing to a removable, nonvolatile optical disk may be provided. In these cases, each drive may be connected to a bus (not shown) by one or more data medium interfaces. Memory 620 may include a computer program product 625 having one or more program modules configured to perform the various methods or acts of the various embodiments of the disclosure.

The communication unit 640 enables communication with other computing devices via a communication medium. Additionally, the functionality of the components of the electronic device 600 may be implemented in a single computing cluster or in multiple computing machines capable of communicating over a communication connection. Thus, the electronic device 600 may operate in a networked environment using logical connections to one or more other servers, a network Personal Computer (PC), or another network node.

The input device 650 may be one or more input devices such as a mouse, keyboard, trackball, etc. The output device 660 may be one or more output devices such as a display, speakers, printer, etc. The electronic device 600 may also communicate with one or more external devices (not shown), such as storage devices, display devices, etc., with one or more devices that enable a user to interact with the electronic device 600, or with any device (e.g., network card, modem, etc.) that enables the electronic device 600 to communicate with one or more other computing devices, as desired, via the communication unit 640. Such communication may be performed via an input/output (I/O) interface (not shown).

According to an exemplary implementation of the present disclosure, a computer-readable storage medium having stored thereon computer-executable instructions, wherein the computer-executable instructions are executed by a processor to implement the method described above is provided. According to an exemplary implementation of the present disclosure, there is also provided a computer program product tangibly stored on a non-transitory computer-readable medium and comprising computer-executable instructions that are executed by a processor to implement the method described above.

Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus, devices, and computer program products implemented according to the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer readable program instructions may be provided to a processing unit of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable medium having the instructions stored therein includes an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various implementations of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The foregoing description of implementations of the present disclosure has been provided for illustrative purposes, is not exhaustive, and is not limited to the implementations disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the various implementations described. The terminology used herein was chosen in order to best explain the principles of each implementation, the practical application, or the improvement of technology in the marketplace, or to enable others of ordinary skill in the art to understand each implementation disclosed herein.

Claims

1. A method for transformation prediction, comprising:

obtaining feature information associated with a target object in a target service demand to be processed and a candidate incentive scheme indicating incentive resources provided for the target object when processing the target service demand;

Determining, using a trained first machine learning model, a base transformation probability for the target object to perform a transformation action in the target service requirement without applying the candidate incentive scheme based on the characteristic information;

determining, using a trained second machine learning model, a conversion probability boost value for the target object to perform the conversion behavior in the target service requirement with the candidate excitation scenario applied based on the feature information; and

Based on the base conversion probability and the conversion probability increase value, a conversion probability is determined that the target object performs the conversion behavior in the target service requirement with the candidate incentive scheme applied.

2. The method of claim 1, wherein the candidate incentive scheme comprises a combination of multiple incentive resources respectively provided for the target object when processing the target service demand with multiple processing categories.

3. The method of claim 2, wherein the second machine learning model comprises a plurality of sub-models corresponding to the plurality of processing categories, respectively, and wherein determining a conversion probability boost value for the target object to perform the conversion behavior in the target service requirement if the candidate incentive scheme is applied comprises:

determining, with each of the plurality of sub-models, a sub-conversion probability increase value for the target object to perform a conversion behavior in the target service requirement with application of the incentive resource of the corresponding processing class based on the characteristic information; and

The conversion probability boost value in the case of applying the candidate excitation scheme is determined based on a plurality of sub-conversion probability boost values determined by the plurality of sub-models.

4. The method of claim 1, wherein determining a conversion probability boost value for the target object to perform the conversion behavior in the target service requirement if the candidate incentive scheme is applied comprises:

with the aid of the second machine learning model,

Extracting from the feature information an intermediate feature representation in case of applying the candidate excitation scheme;

mapping the intermediate feature representation to weight values for the candidate excitation pattern using a mapping function in the second machine learning model; and

The transition probability boost value is determined based on the weight value and the excitation resources indicated by the candidate excitation scheme.

5. The method of claim 4, wherein the mapping function is configured to map the intermediate feature representation to a positive weight value.

6. The method of claim 1, wherein the first machine learning model and the second machine learning model are trained with a plurality of training samples including sample characteristic information of a sample object, a conversion label of the sample object indicating a true conversion condition in which an incentive scheme is not applied or a true conversion condition in which an incentive scheme is applied when processing service requirements of the sample object.

7. The method of claim 6, wherein the training of the first machine learning model and the second machine learning model is further based on the following loss function:

a first loss function, the loss of which is based on, for each training sample:

In case the conversion label indicates a true conversion condition of the non-applied excitation scheme, an error between a predicted base conversion probability determined using the first machine learning model being trained and the true conversion condition of the non-applied excitation scheme, or

In the event that the conversion label indicates a true conversion condition for applying the incentive scheme, utilizing an error between a predicted conversion probability determined by the first machine learning model and the second machine learning model being trained and the true conversion condition for applying the incentive scheme;

A second loss function whose loss is constructed to:

An error between a predicted excitation pattern application determined based on the sample characteristic information and a true excitation pattern application of the sample object using a reference machine learning model being trained.

8. The method of claim 6, wherein the training of the first machine learning model and the second machine learning model is based on the following loss function:

A third loss function, the loss of which is based on, for each training sample: prediction errors of the conversion probability in case that the sample object is not applied with the excitation scheme, and prediction errors applied for the excitation scheme of the sample object.

9. The method of claim 8, wherein the prediction error of the conversion probability comprises:

In the case where the conversion label indicates a true conversion condition to which an excitation scheme is applied, an error between a predicted base conversion probability determined using the first machine learning model being trained and an estimated base conversion probability determined for the sample object when no excitation scheme is applied, the estimated base conversion probability being based on a difference between the true conversion condition to which an excitation scheme is applied and a mean of a plurality of conversion probability elevation values for the plurality of training samples using a second machine learning model; and

Wherein the prediction error applied by the excitation scheme comprises:

10. The method of claim 1, wherein the target service demand comprises a travel demand, the target object comprising a passenger initiating the travel demand or a driver accepting the travel demand; and

Wherein the first machine learning model and the second machine learning model are trained for a passenger side or are trained for a driver side.

11. The method of claim 10, wherein the candidate incentive scheme includes a combination of incentive resources provided for the passenger or driver, respectively, when processing the travel demand with one or more travel patterns.

12. The method of claim 10, wherein the characteristic information associated with the target object comprises at least one of:

Starting point information of the travel demand,

The end point information of the travel demand,

Vehicle supply information of the start point of the travel demand,

Vehicle supply information of the end point of the travel demand,

Historical demand information of the passenger, or

The driver's historical demand information.

13. An apparatus for translation prediction, comprising:

An obtaining module configured to obtain feature information associated with a target object in a target service demand to be processed and a candidate incentive scheme indicating incentive resources provided for the target object when processing the target service demand;

a base prediction module configured to determine a base conversion probability of the target object performing a conversion action in the target service requirement without applying the candidate excitation scheme based on the feature information using a trained first machine learning model;

a boost prediction module configured to determine, based on the feature information, a conversion probability boost value for the target object to perform the conversion behavior in the target service requirement with the candidate excitation scenario applied, using a trained second machine learning model; and

A transformation prediction module configured to determine a transformation probability that the target object performs the transformation behavior in the target service requirement with the candidate incentive scheme applied based on the base transformation probability and the transformation probability elevation value.

14. An electronic device, comprising:

at least one processing unit; and

At least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, which when executed by the at least one processing unit, cause the electronic device to perform the method of any one of claims 1 to 12.

15. A computer readable storage medium having stored thereon a computer program executable by a processor to implement the method of any of claims 1 to 12.

16. A computer program product comprising computer executable instructions which when executed by a processor implement the method of any one of claims 1 to 12.