CN113344590A

CN113344590A - Method and device for model training and complaint rate estimation

Info

Publication number: CN113344590A
Application number: CN202110529815.9A
Authority: CN
Inventors: 陈瑞捷
Original assignee: Beijing Sankuai Online Technology Co Ltd
Current assignee: Beijing Sankuai Online Technology Co Ltd
Priority date: 2021-05-14
Filing date: 2021-05-14
Publication date: 2021-09-03

Abstract

The specification discloses a method and a device for model training and complaint rate prediction, wherein initial model parameters of a complaint rate prediction model of a target sub-scene are obtained by migration according to model parameters of prediction models of other sub-scenes of a service general scene, each sample user executing the service under the target sub-scene is determined, sub-scenes having an association relationship with the sample users are determined from other sub-scenes of the service general scene, each sample data and a label thereof are determined according to user characteristics of each sample user, scene characteristics of the sub-scenes respectively having the association relationship with each sample user and scene characteristics of the target sub-scene, and the complaint rate prediction model is trained based on the prediction results and the labels of each sample data. The method determines training samples based on scene characteristics of other sub-scenes of the total business scene, and determines initial model parameters based on model parameters of a prediction model, so that an accurate complaint rate estimation model can be trained under the condition of less sample amount.

Description

Method and device for model training and complaint rate estimation

Technical Field

The specification relates to the technical field of computers, in particular to a method and a device for model training and complaint rate estimation.

Background

Currently, with the development of computer technology, a service provider can know the intention and attitude of a user for a business in various ways, for example, know the attitude of the user through complaints of the user, and the like. However, excessive complaints occupy a large amount of server resources, and therefore, how to predict the probability of a user complaint before the user complaint and take corresponding measures to avoid the user complaint becomes one of the problems to be solved by the service provider.

In the prior art, a common complaint rate estimation method is determined based on service information of a service executed by a user, where a complaint rate is a probability of complaint performed by the user.

Specifically, first, service information corresponding to a service executed by the user in the service scenario is obtained, for example, service information such as a service serial number, service content, service emergency degree, and a user address is obtained. And secondly, performing Word segmentation on the acquired service information To obtain each Word segmentation result, and determining a Word Vector corresponding To each Word segmentation result through a Word Vector embedding expression (Word To Vector, Word2 Vector) encoding rule. And then, determining a sentence vector corresponding to the service information according to the word vector contained in the service information. And finally, inputting the sentence vector into a pre-trained complaint rate estimation model to obtain the probability that the service corresponding to the service information output by the complaint rate estimation model is complained.

However, in a service scene where a user executes a service less, service information that can be used for training the complaint rate prediction model is less, and the prior art cannot obtain an accurate complaint rate prediction model based on a small amount of service information training.

Disclosure of Invention

The present specification provides a method and an apparatus for model training and complaint rate estimation, which partially solve the above problems in the prior art.

The technical scheme adopted by the specification is as follows:

the training method of the complaint rate estimation model provided by the specification comprises the following steps:

determining initial model parameters of a complaint rate prediction model of a target sub-scene to be trained according to model parameters of pre-trained prediction models of other sub-scenes which belong to the same service general scene as the target sub-scene;

determining sample users corresponding to the historical services according to the historical services executed under the target sub-scene, wherein the service types executed under the sub-scene comprise a specified type and other types;

for each sample user, according to the historical behavior data of the sample user, determining other sub-scenes of the sample user executing the specified type of service from other sub-scenes contained in the total service scene to which the target sub-scene belongs, and taking the other sub-scenes as sub-scenes having an association relationship with the sample user;

determining sample data corresponding to the sample user according to the scene characteristics of the sub-scene having the incidence relation with the sample user, the user characteristics of the sample user and the scene characteristics of the target sub-scene, and determining the label of the sample data according to the historical behavior data of the sample user;

and inputting the complaint rate prediction model to be trained by taking the sample data as input, determining the prediction result of the sample data, adjusting the parameters of the complaint rate prediction model by taking the minimum difference between the prediction result and the label of the sample data as an optimization target, and predicting the probability of the user complaint the target sub-scene by the trained complaint rate prediction model.

Optionally, determining an initial model parameter of a complaint rate prediction model of a target sub-scene to be trained according to a model parameter of a pre-trained prediction model of other sub-scenes belonging to the same service total scene as the target sub-scene, specifically including:

determining historical traffic volume executed under other sub-scenes contained in the total business scene to which the target sub-scene belongs;

determining sub-scenes used for determining initial model parameters of a complaint rate estimation model of the target sub-scene from other sub-scenes according to historical traffic executed under other sub-scenes, and taking the sub-scenes as basic sub-scenes;

and determining initial model parameters of the complaint rate prediction model of the target sub-scene to be trained according to the model parameters of the pre-trained prediction model of the basic sub-scene.

Optionally, the following method is adopted to determine a training sample for training the prediction model of the basic sub-scene: wherein,

determining users corresponding to the historical services as basic users according to the historical services executed under the basic sub-scene;

for each basic user, according to the historical behavior data of the basic user, determining other sub-scenes of the basic user executing the specified type of service from other sub-scenes contained in the service general scene to which the basic sub-scenes belong, and taking the other sub-scenes as sub-scenes which have an association relationship with the basic user;

determining a training sample according to a sub-scene having an incidence relation with the basic user, the user characteristics of the basic user and the scene characteristics of the basic sub-scene, and determining a label of the training sample according to historical behavior data of the basic user.

Optionally, inputting each sample data as an input, inputting the complaint rate prediction model to be trained, and determining a prediction result of each sample data, specifically including:

for each sample data, taking the scene characteristics of the sub-scenes which have an incidence relation with the sample user and the user characteristics of the sample user as input, inputting a user characteristic extraction layer of a complaint rate prediction model to be trained, and determining the historical service characteristics of the sample data;

taking the scene characteristics of the target sub-scene contained in the sample data as input, inputting the scene characteristics extraction layer of the complaint rate prediction model, and determining the scene service characteristics of the sample data;

and inputting the historical service characteristics and the scene service characteristics of the sample data as input into a prediction layer of the complaint rate prediction model, and determining a prediction result of the sample data.

Optionally, determining the label of the sample data according to the historical behavior data of the sample user specifically includes:

taking sample data corresponding to a sample user who executes the specified type of service in the target sub-scene as a positive sample;

and taking the sample data corresponding to the sample user which does not execute the specified type of service in the target sub-scene as a negative sample.

The complaint rate estimation method provided by the specification comprises the following steps:

determining users corresponding to historical services according to the historical services executed under each sub-scene included in the service general scene, wherein the service types executed under the sub-scenes comprise designated types and other types;

for each user, according to the historical behavior data of the user, determining other sub-scenes of the user executing the specified type of service from other sub-scenes contained in the total service scene to which the target sub-scene belongs, and taking the other sub-scenes as sub-scenes having an association relationship with the user;

inputting a pre-trained complaint rate estimation model of the target sub-scene by taking the scene characteristics of the sub-scene having the incidence relation with the user, the user characteristics of the user and the scene characteristics of the target sub-scene as input to obtain the complaint rate of the target sub-scene, wherein the complaint rate is used for carrying out service processing on the user;

the initial model parameters of the complaint rate prediction model are determined according to the model parameters of the prediction models of other sub-scenes belonging to the same service general scene as the target sub-scene, and the complaint rate prediction model is obtained by training at least based on the scene characteristics of the target sub-scene and the scene characteristics of at least part of other sub-scenes belonging to the same service general scene.

The training device of the complaint rate estimation model provided by the specification comprises:

the first determination module is used for determining initial model parameters of a complaint rate estimation model of a target sub-scene to be trained according to model parameters of pre-trained prediction models of other sub-scenes which belong to the same business total scene as the target sub-scene;

the sample user determining module is used for determining sample users corresponding to the historical services according to the historical services executed under the target sub-scene, wherein the service types executed under the sub-scene comprise a designated type and other types;

the sub-scene determining module is used for determining other sub-scenes, which are used by the sample user to execute the specified type of service, from other sub-scenes contained in the total service scene to which the target sub-scene belongs according to historical behavior data of the sample user as the sub-scenes which have an association relationship with the sample user;

the sample determining module is used for determining sample data corresponding to the sample user according to the scene characteristics of the sub-scene having the incidence relation with the sample user, the user characteristics of the sample user and the scene characteristics of the target sub-scene, and determining the label of the sample data according to the historical behavior data of the sample user;

the training module is used for inputting the sample data as input, inputting the complaint rate prediction model to be trained, determining the prediction result of the sample data, adjusting the parameters of the complaint rate prediction model by taking the minimum difference between the prediction result and the label of the sample data as an optimization target, and predicting the probability of the user complaint the target sub-scene by the trained complaint rate prediction model.

The complaint rate estimation device provided by the present specification includes:

the user determining module is used for determining users corresponding to historical services according to the historical services executed under each sub-scene included in the service general scene, wherein the service types executed under the sub-scenes comprise designated types and other types;

the sub-scene determining module is used for determining other sub-scenes, which are used by the user to execute the specified type of service, from other sub-scenes contained in the total service scene to which the target sub-scene belongs according to historical behavior data of the user as the sub-scenes which have an association relationship with the user;

and the prediction module is used for inputting the scene characteristics of the sub-scene having the incidence relation with the user, the user characteristics of the user and the scene characteristics of the target sub-scene, inputting a pre-trained complaint rate prediction model of the target sub-scene to obtain the complaint rate of the target sub-scene, wherein the complaint rate is used for carrying out service processing on the user, the initial model parameters of the complaint rate prediction model are determined according to the model parameters of prediction models of other sub-scenes belonging to the same service general scene as the target sub-scene, and the complaint rate prediction model is obtained by training at least based on the scene characteristics of the target sub-scene and the scene characteristics of at least part of other sub-scenes belonging to the same service general scene.

A computer-readable storage medium provided in this specification stores a computer program, and the computer program, when executed by a processor, implements any one of the above-described methods for training a complaint rate prediction model or a complaint rate prediction method.

The electronic device provided by the present specification includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements any one of the above-mentioned methods for training a complaint rate prediction model or a complaint rate prediction method when executing the program.

The technical scheme adopted by the specification can achieve the following beneficial effects:

in the training method of the complaint rate prediction model provided in this specification, initial model parameters of the complaint rate prediction model of the target sub-scene are determined based on model parameters of prediction models of other sub-scenes of a business general scene to which the target sub-scene belongs, each sample user is determined according to each historical business executed under the target sub-scene, sub-scenes having an association relationship with each sample user are determined from other sub-scenes of the business general scene according to historical behavior data of each sample user, each sample data and a label thereof are determined according to user characteristics of each sample user, scene characteristics of the sub-scenes having an association relationship with each sample user, and scene characteristics of the target sub-scene, and the complaint rate prediction model is trained based on a prediction result and the label of each sample data.

The method can be seen in that the training samples are determined based on the scene characteristics of other sub-scenes in the total business scene, and the initial model parameters are determined based on the model parameters of the prediction model, so that the accurate complaint rate prediction model can be trained under the condition of less sample amount.

Drawings

The accompanying drawings, which are included to provide a further understanding of the specification and are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and together with the description serve to explain the specification and not to limit the specification in a non-limiting sense. In the drawings:

FIG. 1 is a schematic flow chart of a training method of a complaint rate prediction model provided in the present specification;

FIG. 2 is a block diagram of a complaint rate prediction model provided herein;

FIG. 3 is a schematic structural diagram of a user feature extraction layer of a complaint rate prediction model provided in the present specification;

FIG. 4 is a schematic structural diagram of a scene feature extraction layer of a complaint rate prediction model provided in the present specification;

FIG. 5 is a schematic flow chart of a complaint rate estimation method provided in the present specification;

FIG. 6 is a schematic diagram of a training apparatus for a complaint rate prediction model provided in the present specification;

FIG. 7 is a schematic diagram of a complaint rate estimation device provided in the present specification;

fig. 8 is a schematic diagram of an electronic device corresponding to fig. 1 or fig. 5 provided in this specification.

Detailed Description

In order to make the objects, technical solutions and advantages of the present disclosure more clear, the technical solutions of the present disclosure will be clearly and completely described below with reference to the specific embodiments of the present disclosure and the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present disclosure, and not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present specification without any creative effort belong to the protection scope of the present specification.

In the prior art, a complaint rate prediction model needs to be obtained by training based on a sentence vector corresponding to service information of each service executed in a service scene, so that when a user executes the service in the service scene, a method for determining the complaint rate of the user for the executed service can be determined based on the complaint rate prediction model. The method is characterized in that initial model parameters of the complaint rate prediction model of the service scene are determined based on model parameters of the complaint rate prediction model of other scenes belonging to the same service general scene as the service scene, and the complaint rate prediction model of the service scene is obtained through training based on the initial model parameters and training samples. Under the condition that the user does not execute the service in the service scene, the complaint rate of the user to the service scene can be predicted, and the condition that an accurate complaint rate prediction model cannot be determined if the service information of each historical service executed by the user in the service scene is less is avoided.

The technical solutions provided by the embodiments of the present description are described in detail below with reference to the accompanying drawings.

Fig. 1 is a schematic flow chart of a training method of a complaint rate prediction model provided in this specification, specifically including the following steps:

s100: according to model parameters of pre-trained prediction models of other sub-scenes which belong to the same business total scene as the target sub-scene, determining initial model parameters of a complaint rate prediction model of the target sub-scene to be trained.

Generally, a service provider can predict users through a complaint rate prediction model, determine the complaint rate of each user for a current service scene, and take relevant measures, generally, the complaint rate prediction model is obtained by a server for training the model and training in advance based on a training sample, and the specification provides a new method for training the complaint rate prediction model, and similarly, the server for training the model can execute the process of training the complaint rate prediction model.

In this specification, a user may execute services corresponding to each service type, such as a complaint service, a consultation service, and the like, under each sub-scene included in one service overall scene, and then, for each sub-scene included in the service overall scene, a complaint rate corresponding to the user may be determined based on a historical service executed by the user in the scene. However, in sub-scenarios with less historical services, an accurate complaint rate estimation model cannot be obtained through training according to less service information, and for each user, it can be considered that the services executed by the user under each sub-scenario included in the same service general scenario have relevance, for example, if the same service general scenario includes a pop-up window scenario and a payment scenario, if the user executes a complaint service under the pop-up window scenario, it can be considered that the probability of the user executing the complaint service under the payment scenario is also higher. Therefore, the prediction model of the complaint rate of the target sub-scene can be determined based on the prediction models of other sub-scenes belonging to the same business general scene as the target sub-scene.

Specifically, after the target sub-scene is determined, the server may obtain model parameters of a pre-trained prediction model of other sub-scenes belonging to the same business general scene as the target sub-scene, and determine initial model parameters of a complaint rate prediction model of the target sub-scene to be trained according to the obtained model parameters of the prediction model.

It should be noted that, in the specific method for determining the initial model parameters of the complaint rate prediction model according to the model parameters of the prediction model, the method for compiling the model parameters of the prediction model into the general parameter space and determining the initial model parameters of the complaint rate prediction model according to the model parameters of the general parameter space may be adopted.

S102: and determining sample users corresponding to the historical services according to the historical services executed under the target sub-scene, wherein the service types executed under the sub-scene comprise a specified type and other types.

The training model can be divided into a sample generation stage and a training model stage, and in the sample generation stage, samples for training the model can be determined according to model requirements and training requirements. In this specification, the server may determine a training sample for training the complaint rate prediction model, and since the complaint rate prediction model is trained based on user information or the like of a user who has performed a service in the target sub-scenario, the server may determine a sample user who has performed a service in the target sub-scenario.

Specifically, the server may obtain each historical service executed in the target sub-scenario, and determine, for each historical service, a user executing the historical service according to the service information of the historical service, as a sample user corresponding to the historical service.

In addition, as described above, the user may perform services corresponding to each service type, such as a complaint service, a consultation service, and the like, in sub-scenes. In the present specification, it is necessary to determine the complaint rate of the user about the target sub-scenario, and therefore, the sub-scenario in which the user has performed the specified type of service can be determined in the subsequent step. In each sub-scenario, a user may perform a specified type of service as well as other types of services. The service of the specified type is a service that needs to be predicted, that is, a service that needs to be predicted whether a user will execute or not. For example, if it is predicted whether the User clicks the pushed advertisement, the specified type of service is an advertisement pushing service, or if it is predicted whether the User uploads User original Content (UCC), the specified type of service is a User publishing service, and so on. The service of the specified type can be set as required, and for a scene in which the complaint behavior of the user needs to be predicted in the specification, the service of the specified type is a service related to the complaint rate, such as a complaint service and a reporting service. Of course, services such as reward services may be used to determine the complaint rate in the reverse direction. The specific type can be set according to the needs, and the specification does not limit the specific type.

S104: and for each sample user, according to the historical behavior data of the sample user, determining other sub-scenes of the sample user, which have executed the specified type of service, from other sub-scenes contained in the total service scene to which the target sub-scene belongs, as the sub-scenes which have an association relationship with the sample user.

In this specification, as described above, it can be considered that the same user has a correlation between the services executed by the sub-scenarios included in one total service scenario, and therefore, in model training, the training sample can be determined based on the other sub-scenarios included in the total service scenario and executed by the user of each sample.

Specifically, the server may determine, for each sample user determined in step S102, historical behavior data of the sample user, where the historical behavior data includes service information of each service executed by the sample user under each sub-scene included in the total service scene to which the target sub-scene belongs. And according to the historical behavior data of the user, the server can determine other sub-scenes of the specified type of service executed by the sample user from other sub-scenes contained in the total scene of the service, and takes the sub-scenes as the sub-scenes which have an association relationship with the sample user.

In addition, the number of the sub-scenes determined to have the association relationship with the sample user may be multiple, that is, the sample user has performed the specified type of service in the multiple sub-scenes included in the total scene of the service. Taking a service of a specified type as an example of a complaint service, assuming that a total service scene includes a payment scene, a pop-up window scene and a preferential scene, a target sub-scene is a payment scene, and the sample user executes the complaint service in both the payment scene and the pop-up window scene, it can be determined that the sub-scenes having an association relationship with the sample user include the payment scene and the pop-up window scene.

It should be noted that the historical behavior data may also only include service information of a service of a specified type that is executed by the sample user under each sub-scene included in the total service scene to which the target sub-scene belongs, and specific content of the historical behavior data may be set according to needs, which is not limited in this specification.

S106: according to the scene characteristics of the sub-scene having the incidence relation with the sample user, the user characteristics of the sample user and the scene characteristics of the target sub-scene, sample data corresponding to the sample user is determined, and according to the historical behavior data of the sample user, the label of the sample data is determined.

In one or more embodiments provided in this specification, when training a complaint rate prediction model, if a training sample is determined based on only service information of a service executed by a sample user in a target sub-scene, then when subsequently using the complaint rate prediction model to perform complaint rate prediction, if the user has not executed a service in the target sub-scene, then the complaint rate of the user with respect to the target sub-scene cannot be determined. For example, assuming that the target sub-scene is a pop-up window scene, if the user has not executed a service in the pop-up window scene, the complaint rate of the user with respect to the pop-up window scene cannot be determined.

However, if the training sample is determined based on the user characteristics of the sample users, the scene characteristics of the sub-scenes respectively associated with the sample users, and the scene characteristics of the target sub-scene, based on the complaint rate prediction model obtained by training, not only the complaint rate of each user corresponding to the historical service executed in the target sub-scene with respect to the target sub-scene but also the complaint rate of each user corresponding to each historical service executed in other sub-scenes included in the total scene of the service with respect to the target sub-scene can be determined. For example, assuming that the total business scene includes a payment scene, a pop-up window scene and a preferential scene, and the target sub-scene is the payment scene, based on the user characteristics of each sample user, the scene characteristics of each sub-scene respectively having an association relationship with each sample user, and the complaint rate prediction model trained from the scene characteristics of the target sub-scene, not only can the complaint rate of the user who has executed the business in the payment scene with respect to the payment scene be determined, but also the complaint rate of the user who has executed the business in the pop-up window scene and/or the preferential scene with respect to the payment scene can be determined.

Specifically, the server may first obtain, for each sample user, user information of the sample user, scene information of a sub-scene having an association relationship with the sample user, and scene information of the target sub-scene, so as to determine a training sample subsequently. The user information of each sample user may include user image information such as an age interval of the sample user, a gender of the sample user, and the scene information may include scene general information and scene specific information. The scene general information may include a flow of a scene, whether each user performs a specified type of service in the scene, a channel through which each user performs the specified type of service, and the like. The scenario-specific information may include reasons why the user performed a specified type of service, characteristics specific to the scenario, and the like. Taking a pop-up window scene as an example, the reason for the user to execute the specified type of service may include frequent pop-up window, irrelevant pop-up window content, and the like, and the characteristic of the scene may include the number of pop-up windows, the number of user clicks, and the like. Of course, the contents of the specific user information and the scene information may be set as required, and this specification does not limit this.

Then, the server may perform feature extraction on the user information of the sample user, the scene information of the sub-scene having the association relationship with the sample user, and the scene information of the target sub-scene, respectively, to obtain the user feature of the sample user, the scene feature of the sub-scene having the association relationship with the sample user, and the scene feature of the target sub-scene, and use the determined user feature of the sample user, the scene feature of the sub-scene having the association relationship with the sample user, and the scene feature of the target sub-scene as sample data corresponding to the sample user.

And finally, the server can judge whether the sample user executes the specified type of service in the target sub-scene according to the historical behavior data of the sample user, and determine the label of the sample data according to the judgment result.

Further, in this specification, when determining the label of each sample data, the server may use, as a positive sample, the sample data corresponding to the sample user who has executed the specified type of service in the target sub-scene, where the label of the positive sample indicates that the specified type of service has been executed. And taking sample data corresponding to a sample user which does not execute the specified type of service in the target sub-scene as a negative sample, wherein the label of the negative sample represents that the specified type of service is not executed.

In addition, the complaint rate determined by the complaint rate estimation model is the probability of the user for the target sub-scene complaint, and the interval of the probability is generally 0-100%, so that the label of the positive sample can be determined to be 100%, and the label of the negative sample can be 0, so that loss can be calculated based on the estimation result and the label of each sample data. For example, if the prediction for a positive sample is 60%, then the gap between the prediction for the positive sample and the label may be determined to be 40%, and then the loss may be determined based on the 40%. Of course, the numerical values and the like of the specific labels can be set as required, and the numerical values and the like can be used for representing the complaint rate of the sample user on the target sub-scene, and the description does not limit the complaint rate.

In this specification, the manner of extracting the features of each piece of information may be set as needed, and when sample data is determined, user information of each sample user, scene information of a sub-scene associated with each sample user, and scene information of the target sub-scene may be combined to determine each piece of sample data.

S108: and inputting the complaint rate prediction model to be trained by taking the sample data as input, determining the prediction result of the sample data, adjusting the parameters of the complaint rate prediction model by taking the minimum difference between the prediction result and the label of the sample data as an optimization target, and predicting the probability of the user complaint the target sub-scene by the trained complaint rate prediction model.

In one or more embodiments provided in this specification, after determining each training sample and its label, the server may input each sample data as an input, input the complaint rate prediction model to be trained, determine a prediction result of each sample data, determine a loss function based on the prediction result and the label of each sample data, minimize the loss function as an optimization target, and adjust parameters of the complaint rate prediction model.

Specifically, for each sample data, the server may first input, to a user feature extraction layer of the complaint rate prediction model to be trained, a user feature of a sample user included in the sample data and a scene feature of a sub-scene having an association relationship with the sample user as inputs, and determine a historical service feature of the sample data.

Then, the server may input the scene characteristics of the target sub-scene included in the sample data as input to a scene characteristic extraction layer of the complaint rate prediction model, and determine the scene service characteristics of the sample data.

And finally, the server can input the historical service characteristics and the scene service characteristics of the sample data into a prediction layer of the complaint rate prediction model to obtain a prediction result of the sample data. As shown in fig. 2.

Fig. 2 is a structural diagram of a complaint rate prediction model provided in this specification, where a user characteristic is a user characteristic of a sample user, other sub-scene characteristics are scene characteristics of sub-scenes having an association relationship with the sample user, and a target sub-scene characteristic is a scene characteristic of a target sub-scene. And as can be seen, the server takes the user characteristics of the sample user and the scene characteristics of the sub-scenes having the association relationship with the sample user as input, inputs the input into the user characteristic extraction layer of the complaint rate prediction model, and determines the historical service characteristics of the sample data. And inputting the scene characteristics of the target sub-scene as input into a scene characteristic extraction layer of the complaint rate prediction model, and determining the scene service characteristics of the sample data. And then inputting the historical service characteristics and the scene service characteristics of the sample data as input into a prediction layer of the complaint rate prediction model to obtain a prediction result of the sample data.

The estimation result of the determined sample data is determined comprehensively based on two dimensions, namely the user dimension and the scene dimension. From the user dimension, the probability of executing the specified type of service in the target sub-scene is different for different types of users, and therefore, different pre-estimated results of the different types of users for the target sub-scene can be determined based on the user characteristics and the scene characteristics of the target sub-scene. From the scene dimension, the sub-scenes having an association relationship with each user are actually the sub-scenes in which each user has executed a service of a specified type in the total scene of the service, and based on the scene features of the sub-scenes having an association relationship with each user and the scene features of the target sub-scenes, the probability that the user executes the service of the instruction type in the target sub-scenes, that is, the estimation result, can be predicted.

Further, the structure of the user feature extraction layer of the complaint rate prediction model can be as shown in fig. 3, where the user feature is a user feature of a sample user, the scenes 1 to n are sub-scenes having an association relationship with the sample user, and the scene features of the scenes 1 to n are scene features of sub-scenes having an association relationship with the sample user. Therefore, the server can firstly input the user characteristics of the sample user and the scene characteristics of the scenes 1 to n into the embedding layer of the user characteristic extraction layer respectively to obtain the user embedding vectors of the sample user and the embedding vectors of the scenes 1 to n output by the embedding layer. Then, the server can fuse the embedded vectors of the scenes 1 to n, and input the fusion result into the pooling layer of the user feature extraction layer to obtain the historical scene vector of the sample user output by the pooling layer. Then, the server can fuse the user embedded vector of the sample user and the historical scene vector of the sample user, input the fusion result into the activation layer of the user feature extraction layer, and obtain the historical service feature of the sample data output by the activation layer through an activation function. The number of the sub-scenes which are input to the user feature extraction layer and have the association relationship with the sample user may be only one or multiple, and the number of the scene features which are input to the sub-scenes which are input to the user feature extraction layer and have the association relationship with the sample user, that is, the number of n, may be set as required, which is not limited in this specification.

Further, the scene feature extraction layer of the complaint rate prediction model may be as shown in fig. 4, where scene features 1 to m are each scene feature of the target sub-scene, the server may input the scene features 1 to m of the target sub-scene into the embedding layer, determine scene embedding vectors 1 to m of the target sub-scene, fuse the scene embedding vectors 1 to m, input a fusion result thereof into the pooling layer, and determine a scene service feature of the target sub-scene output by the pooling layer as a scene service feature of the sample feature. The number of the scene features of the target sub-scene input to the scene feature extraction layer may be only one, or may also be multiple, and the specific number of the scene features of the target sub-scene input to the scene feature extraction layer, that is, the number of m, may be set as required, which is not limited in this specification.

In addition, the prediction layer of the complaint rate prediction model can be specifically a function determined by the complaint rate

And determining the estimated result of the sample data. Wherein, V_uFor historical traffic characteristics, V_CFor scene service characteristics, u is a user, C is a target sub-scene, and logic () is a logic function.

In the present specification, the feature fusion may be determined by processing each feature such as concatenation and addition. It is a mature technology to specifically fuse each feature, and this description is not repeated here.

The training method based on the complaint rate prediction model shown in FIG. 1 determines the initial model parameters of the complaint rate prediction model of the target sub-scene to be trained based on the model parameters of the prediction models of other sub-scenes belonging to the same service general scene as the target sub-scene, determining each sample user according to each historical service executed under the target sub-scene, determining the historical behavior data of each sample user from other sub-scenes of the total service scene to which the target sub-scene belongs, sub-scenarios having an association with sample users are determined, and for each sample user, determining sample data according to the user characteristics of the sample user, the scene characteristics of the sub-scene having the incidence relation with the sample user and the scene characteristics of the target sub-scene, and determining the label of the sample data according to the historical behavior data of the sample user, and then training the complaint rate estimation model based on the estimation result and the label of each sample data. The method determines training samples based on scene characteristics of other sub-scenes in the total business scene, and determines initial model parameters based on model parameters of a prediction model, so that an accurate complaint rate estimation model can be trained under the condition of less sample amount.

Furthermore, in the field of machine learning, a more accurate model can be obtained based on a large number of samples, so that in order to ensure the accuracy of the complaint rate prediction model obtained through training, when the prediction models of other sub-scenes for determining the initial model parameters of the complaint rate prediction model are determined, the sub-scene with the largest historical traffic can be determined from the total traffic scene according to the historical traffic of each sub-scene, and the determined prediction model of the sub-scene is used as the prediction model for determining the initial model parameters of the complaint rate prediction model.

Further, the server may first determine the historical traffic volume performed in other sub-scenes included in the total scene of the business to which the target sub-scene belongs. The server may then sort the sub-scenes in order of decreasing historical traffic. And finally, determining the sub-scenes used for determining the initial model parameters of the complaint rate prediction model from all the sub-scenes according to the sequence, and taking the sub-scenes as basic sub-scenes. For example, the sub-scene with the highest historical traffic volume is selected as the basic sub-scene, or the sub-scene with the second highest historical traffic volume is selected as the basic sub-scene, and how to determine the basic sub-scene according to the ranking may be set as needed, which is not limited in this specification.

In this specification, the prediction model of the base sub-scenario may be trained by a server that trains the model. When training the prediction model, the server may first determine each historical service executed in the basic sub-scenario, and determine, according to each determined historical service, a user corresponding to each historical service as each basic user. Secondly, for each basic user, according to the historical behavior data of the basic user, determining other sub-scenes of the basic user, which have executed the specified type of service, from the target sub-scenes, except the basic sub-scene, included in the total service scene to which the basic sub-scene belongs, as the sub-scenes which have an association relationship with the basic user. Then, the server may determine a training sample according to the sub-scenario having an association relationship with the base user, the user characteristics of the base user, and the scenario characteristics of the base sub-scenario, and determine a label of the training sample according to the historical behavior data of each base user. And finally, inputting each training sample into a prediction model to be trained, determining the prediction result of each training sample, and training the prediction model based on the prediction result and the label of each training sample.

Based on the training method of the complaint rate prediction model shown in fig. 1, the present specification further provides a complaint rate prediction method, as shown in fig. 5.

Fig. 5 is a schematic flow chart of a complaint rate estimation method provided in this specification, including:

s200: and determining users corresponding to the historical services according to the historical services executed under the sub-scenes included in the service general scene, wherein the service types executed under the sub-scenes comprise a specified type and other types.

In one or embodiment provided in this specification, it is possible to predict a complaint rate of each user who has performed a service under each sub-scene included in a total service scene to which a target sub-scene belongs, and therefore, the server may determine a user corresponding to each historical service according to each historical service performed under each sub-scene included in the total service scene.

S202: and for each user, according to the historical behavior data of the user, determining other sub-scenes in which the user executes the specified type of service from other sub-scenes contained in the total service scene to which the target sub-scene belongs, and taking the other sub-scenes as the sub-scenes which have an association relationship with the user.

S204: inputting a pre-trained complaint rate estimation model of the target sub-scene by taking the scene characteristics of the sub-scene having the incidence relation with the user, the user characteristics of the user and the scene characteristics of the target sub-scene as input to obtain the complaint rate of the target sub-scene, wherein the complaint rate is used for carrying out service processing on the user; the initial model parameters of the complaint rate prediction model are determined according to the model parameters of the prediction models of other sub-scenes belonging to the same service general scene as the target sub-scene, and the complaint rate prediction model is obtained by training at least based on the scene characteristics of the target sub-scene and the scene characteristics of at least part of other sub-scenes belonging to the same service general scene.

In the present specification, the first one or more embodiments are provided, because the complaint rate prediction model used in the present specification is obtained by training based on the model parameters of the prediction models of other sub-scenes belonging to the same business general scene as the target sub-scene, the scene characteristics of at least some other sub-scenes belonging to the same business general scene, the user characteristics of the user, and the like, when complaint rate prediction is performed, the user characteristics of each user, the scene characteristics of the target sub-scene, and the scene characteristics of the sub-scenes having an association relationship with each user can be obtained, and the complaint rate of each user with respect to the target sub-scene is determined.

For a specific method for determining the user characteristics of each user, the scene characteristics of the target sub-scene, and the scene characteristics of each sub-scene having an association relationship with each user, and a method for determining the complaint rate through the complaint rate estimation model, reference may be made to the contents of steps S104 to S108, which are not described herein again.

Based on the same idea, the present specification further provides a training device and a complaint rate prediction device for a complaint rate prediction model, as shown in fig. 6 or 7.

Fig. 6 is a schematic diagram of a training device of a complaint rate prediction model provided in this specification, including:

the first determining module 300 is configured to determine, according to model parameters of a pre-trained prediction model of other sub-scenes belonging to the same business general scene as the target sub-scene, initial model parameters of a complaint rate prediction model of the target sub-scene to be trained.

A sample user determining module 302, configured to determine, according to each historical service executed in the target sub-scenario, a sample user corresponding to each historical service, where the service type executed in the sub-scenario includes a specified type and other types.

And the sub-scene determining module 304 is configured to determine, for each sample user, according to the historical behavior data of the sample user, from other sub-scenes included in the total business scene to which the target sub-scene belongs, other sub-scenes in which the sample user has executed a specified type of business, as sub-scenes having an association relationship with the sample user.

The sample determining module 306 is configured to determine sample data corresponding to the sample user according to the scene features of the sub-scene having the association relationship with the sample user, the user features of the sample user, and the scene features of the target sub-scene, and determine a tag of the sample data according to the historical behavior data of the sample user.

The training module 308 is configured to input each sample data as an input, input the complaint rate prediction model to be trained, determine a prediction result of each sample data, adjust parameters of the complaint rate prediction model by using the minimum difference between the prediction result and the label of each sample data as an optimization target, and predict the probability that the user complains the target sub-scene by the trained complaint rate prediction model.

Optionally, the first determining module 300 is specifically configured to determine historical traffic volumes executed under other sub-scenes included in a total business scene to which the target sub-scene belongs, determine, according to the historical traffic volumes executed under the other sub-scenes, a sub-scene used for determining initial model parameters of a complaint rate prediction model of the target sub-scene from the other sub-scenes, and determine, as a basic sub-scene, initial model parameters of a complaint rate prediction model of the target sub-scene to be trained according to model parameters of a pre-trained prediction model of the basic sub-scene.

Optionally, the first determining module 300 is specifically configured to determine, according to each historical service executed in a basic sub-scenario, a user corresponding to each historical service, as each basic user, and according to historical behavior data of the basic user, determine, from other sub-scenarios included in a total scenario of services to which the basic sub-scenario belongs, other sub-scenarios in which the basic user has executed a specified type of service, as sub-scenarios having an association relationship with the basic user, determine, according to a sub-scenario having an association relationship with the basic user, a user characteristic of the basic user, and a scenario characteristic of the basic sub-scenario, a training sample, and determine a label of the training sample according to historical behavior data of the basic user.

Optionally, the training module 308 is specifically configured to, for each sample data, input a user feature extraction layer of a complaint rate prediction model to be trained, to determine a historical service feature of the sample data, input a scene feature of the target sub-scene included in the sample data, input the scene feature extraction layer of the complaint rate prediction model, determine a scene service feature of the sample data, input the historical service feature and the scene service feature of the sample data as inputs, and input the historical service feature and the scene service feature of the sample data to the prediction layer of the complaint rate prediction model, to determine a prediction result of the sample data.

Optionally, the sample determining module 306 is specifically configured to use, as a positive sample, sample data corresponding to a sample user who has executed the specified type of service in the target sub-scenario, and use, as a negative sample, sample data corresponding to a sample user who has not executed the specified type of service in the target sub-scenario.

Fig. 7 is a schematic diagram of a complaint rate estimation device provided in the present specification, including:

a user determining module 400, configured to determine, according to each historical service executed in each sub-scene included in a service general scene, a user corresponding to each historical service, where a service type executed in the sub-scene includes a specified type and other types;

a sub-scenario determining module 402, configured to determine, for each user, according to historical behavior data of the user, from other sub-scenarios included in the total business scenario to which the target sub-scenario belongs, another sub-scenario in which the user has executed a specified type of business, as a sub-scenario having an association relationship with the user.

The prediction module 404 is configured to input a scene feature of a sub-scene having an association relationship with the user, a user feature of the user, and a scene feature of the target sub-scene as inputs, input a pre-trained complaint rate prediction model of the target sub-scene, and obtain a complaint rate of the target sub-scene, where the complaint rate is used for performing service processing on the user, and an initial model parameter of the complaint rate prediction model is determined according to a model parameter of a prediction model of another sub-scene belonging to the same service general scene as the target sub-scene, and is trained to obtain the complaint rate prediction model based on at least the scene feature of the target sub-scene and scene features of at least some other sub-scenes belonging to the same service general scene.

The present specification also provides a computer-readable storage medium storing a computer program, which is operable to execute at least one of the above-described method for training a complaint rate prediction model provided in fig. 1 and the above-described method for complaint rate prediction provided in fig. 5.

This specification also provides a schematic block diagram of the electronic device shown in fig. 8. As shown in fig. 8, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may include hardware required for other services. The processor reads a corresponding computer program from the nonvolatile memory to the memory and then runs the computer program to implement the training method of the complaint rate prediction model shown in fig. 1 and the complaint rate prediction method shown in fig. 5. Of course, besides the software implementation, the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.

In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually making an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as abel (advanced Boolean Expression Language), ahdl (alternate Hardware Description Language), traffic, pl (core universal Programming Language), HDCal (jhdware Description Language), lang, Lola, HDL, laspam, hardward Description Language (vhr Description Language), vhal (Hardware Description Language), and vhigh-Language, which are currently used in most common. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.

The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.

The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The above description is only an example of the present specification, and is not intended to limit the present specification. Various modifications and alterations to this description will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.

Claims

1. A training method of a complaint rate prediction model is characterized by comprising the following steps:

2. The method of claim 1, wherein determining initial model parameters of a complaint rate prediction model of a target sub-scene to be trained according to model parameters of pre-trained prediction models of other sub-scenes belonging to the same business general scene as the target sub-scene specifically comprises:

3. The method of claim 2, wherein training samples for training a predictive model of the base sub-scenario are determined using the following method: wherein,

4. The method of claim 1, wherein the step of inputting the sample data as input into the complaint rate prediction model to be trained and determining the prediction result of the sample data comprises:

5. The method of claim 1, wherein determining the label of the sample data according to the historical behavior data of the sample user specifically comprises:

6. A complaint rate prediction method, comprising:

7. A training device for a complaint rate prediction model is characterized by comprising:

8. A complaint rate prediction apparatus, characterized in that the apparatus comprises:

9. A computer-readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1 to 5 or 6.

10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1 to 5 or 6 when executing the program.