CN115935031A - User screening method and device, electronic equipment and storage medium - Google Patents

Abstract

The invention discloses a user screening method, a user screening device, electronic equipment and a storage medium, wherein the method comprises the following steps: acquiring a user portrait of each target user, and screening candidate users from each target user according to the user portrait; acquiring the push value of each candidate user and the intention probability aiming at the specified category; the push value and the intention probability are both obtained by analyzing the user portrait; and screening users for pushing the information under the specified category from the candidate users according to the pushing value and the intention probability. After candidate users are screened out from target users, the push value and the intention probability aiming at the appointed class of the candidate users are obtained, and the candidate users are further screened according to the push value and the intention probability, so that the push precision of the crowd is improved, and the push coverage and the push effectiveness are further improved. And the intention probability of the user aims at the category to which the information to be pushed belongs, so that the targeted final user is stronger, and the crowd pushing precision can be further improved.

Description

User screening method and device, electronic equipment and storage medium

Technical Field

The invention relates to the technical field of computers, in particular to a user screening method and device, electronic equipment and a storage medium.

Background

Currently, in the similar population expansion scheme, similarity between candidate users (belonging to users other than the seed user) and a seed user set is calculated, and a certain number of candidate users are selected as a similar population package of the seed user set according to a sequence from high similarity to low similarity.

However, the screening precision of the similar crowd packets determined based on the similarity manner is not high, so that the information pushing effect is reduced.

Disclosure of Invention

The present invention provides a user screening method, an apparatus, an electronic device and a storage medium for overcoming the above-mentioned deficiencies in the prior art, and the object is achieved by the following technical solutions.

The first aspect of the present invention provides a user screening method, including:

acquiring a user portrait of each target user, and screening candidate users from each target user according to the user portrait;

acquiring the push value of each candidate user and the intention probability aiming at the specified category; the push value and the intention probability are both obtained by analyzing the user portrait;

and screening users for pushing the information under the appointed category from the candidate users according to the pushing value and the intention probability.

In some embodiments of the present application, the screening candidate users from respective target users according to user figures includes:

for each target user, predicting a push probability according to a user portrait of the target user; and screening candidate users from all target users according to the pushing probability.

In some embodiments of the present application, said predicting a push probability from a user representation of the target user comprises:

extracting a first specified type of user feature based on a user representation of the target user; inputting the user characteristics into a trained first classification model, and predicting and outputting the pushing probability of a target user according to the user characteristics by the first classification model.

In some embodiments of the present application, the training process for the first classification model comprises:

extracting user features of a first specified type as positive sample features based on the user portrait of the ordered user; extracting a first specified type of user features as negative sample features based on a user portrait of a user who has not placed a order and has a liveness higher than a first threshold; and training the constructed first classification model by using the positive sample characteristics and the negative sample characteristics until convergence.

In some embodiments of the present application, the filtering candidate users from the target users according to the push probability includes:

and selecting the target users with the push probability larger than a second threshold value from all the target users as candidate users.

In some embodiments of the present application, the obtaining the push value of each candidate user includes:

for each candidate user, extracting a second specified type of user feature based on a user representation of the candidate user; inputting the user characteristics of the second specified type into different regression models respectively so that each regression model can predict the sub-push value of the candidate user according to the user characteristics; and determining the push value of the candidate user according to the sub-push values output by each regression model.

In some embodiments of the present application, the training process for each regression model comprises:

extracting a user characteristic of a second specified type as a positive sample characteristic based on the user portrait of the order-issued user, and acquiring the average transaction amount of the order-issued user as a sample label corresponding to the positive sample characteristic; extracting a user characteristic of a second specified type as a negative sample characteristic based on the user portrait of the user who does not give the order, and taking a first preset value as a sample label corresponding to the negative sample characteristic; and respectively training each regression model by utilizing the positive sample characteristics and the corresponding sample labels thereof and the negative sample characteristics and the corresponding sample labels thereof until convergence.

In some embodiments of the present application, after each regression model is trained separately using positive sample features and their corresponding sample labels, negative sample features and their corresponding sample labels, the method further comprises:

respectively inputting the negative sample characteristics of the users who do not order into each trained regression model, and predicting the sub-push value of the users who do not order according to the negative sample characteristics by each trained regression model; determining the push value of the users who do not leave the order according to the sub-push value output by each regression model, and updating the sample label corresponding to the negative sample feature by using the push value; and (4) retraining each trained regression model respectively by using the negative sample characteristics and the corresponding updated sample labels until convergence.

In some embodiments of the present application, the obtaining the push value of each candidate user includes:

for each candidate user, if the candidate user generates ordering behavior, acquiring the average transaction amount of the candidate user as a push value; and if the candidate user does not generate ordering behavior, taking the second preset value as a pushing value.

In some embodiments of the present application, the obtaining an intention probability of each candidate user for the specified category includes:

for each candidate user, extracting a first specified type of user feature based on a user portrait of the candidate user; and inputting the user characteristics of the first specified type into a second classification model corresponding to the specified category so as to predict the intention probability of the candidate user for the specified category according to the user characteristics by the second classification model.

In some embodiments of the present application, the training process for the second classification model comprises:

acquiring the ordered user under the specified category, and extracting a user feature of a first specified type as a positive sample feature based on the acquired user portrait of the ordered user; extracting user features of a first specified type as negative sample features based on user figures of users who have not made orders; and training the constructed second classification model by using the positive sample characteristics and the negative sample characteristics until convergence.

In some embodiments of the present application, the screening users from candidate users for pushing information under the specified category according to the pushing value and the intention probability includes:

and selecting candidate users with push values higher than a third threshold value and intention probabilities higher than a fourth threshold value from the candidate users as the users for pushing the information under the specified category.

A second aspect of the present invention provides a user screening apparatus, including:

the first screening module is used for acquiring the user portrait of each target user and screening candidate users from each target user according to the user portrait;

the acquisition module is used for acquiring the push value of each candidate user and the intention probability aiming at the specified category; the push value and the intention probability are obtained by analyzing the user portrait;

and the second screening module is used for screening the users for pushing the information under the specified category from the candidate users according to the pushing value and the intention probability.

A third aspect of the present invention proposes an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the method according to the first aspect when executing the program.

A fourth aspect of the present invention proposes a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, carries out the steps of the method according to the first aspect as described above.

Based on the user screening method and device in the first and second aspects, the invention has at least the following beneficial effects or advantages:

after candidate users are screened out from target users, the push value and the intention probability for the appointed class of the users are obtained, and the candidate users are further screened according to the push value and the intention probability, so that the push precision of the crowd is improved, and the push coverage and effectiveness are further improved. And because the intention probability of the user aims at the category to which the information to be pushed belongs, the targeted performance of the screened final user is stronger, and the pushing precision of the crowd can be further improved.

Drawings

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

FIG. 1 is a flowchart illustrating an embodiment of a user screening method according to an exemplary embodiment of the present invention;

FIG. 2 is a schematic diagram illustrating a pushed value prediction process of a candidate user according to the embodiment shown in FIG. 1;

FIG. 3 is a schematic diagram illustrating a process of probability prediction of intention of a candidate user for a specified category according to the embodiment of FIG. 1;

fig. 4 is a schematic structural diagram illustrating a user screening apparatus according to an exemplary embodiment of the present invention;

FIG. 5 is a diagram illustrating a hardware configuration of an electronic device according to an exemplary embodiment of the present invention;

fig. 6 is a schematic diagram illustrating a structure of a storage medium according to an exemplary embodiment of the present invention.

Detailed Description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. The following description refers to the accompanying drawings in which the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the invention, as detailed in the appended claims.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

It is to be understood that although the terms first, second, third, etc. may be used herein to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present invention. The word "if," as used herein, may be interpreted as "at … …" or "at … …" or "in response to a determination," depending on the context.

The quality of the existing similar crowd packets screened based on the similarity mode needs to be evaluated by acquiring behaviors generated by the users after the users are online and information is pushed, so that the loss is brought to an information pushing party due to the problem of low screening precision.

In order to solve the technical problem, the application provides a user screening method, namely, user figures of all target users are obtained, candidate users are screened from all target users according to the user figures, then the pushing value and the intention probability of each candidate user for the appointed category are obtained, the pushing value and the intention probability are obtained by analyzing the user figures, and users used for pushing information under the appointed category are further screened from the candidate users according to the pushing value and the intention probability.

The technical effects which can be achieved based on the technical scheme described above are as follows:

after candidate users are screened out from target users, the push value and the intention probability for the appointed class of the users are obtained, and the candidate users are further screened according to the push value and the intention probability, so that the push precision of the crowd is improved, and the push coverage and effectiveness are further improved. And the intention probability of the user aims at the category to which the information to be pushed belongs, so that the targeted final user is stronger, and the pushing precision of the crowd can be further improved.

In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.

The first embodiment is as follows:

fig. 1 is a flowchart illustrating an embodiment of a user screening method according to an exemplary embodiment of the present invention, where the user screening method may be applied to any electronic device capable of being networked, as shown in fig. 1, and the user screening method includes the following steps:

step 101: and acquiring the user portrait of each target user, and screening candidate users from each target user according to the user portrait.

In this embodiment, the target user refers to a user to be expanded selected by the information presenter. The user portrait is obtained by the background server of the APP installed in the device performing ETL (Extract Transform Load) processing according to the user natural attribute information and the user social attribute information.

The user natural attribute information refers to personal information filled when a user logs in and registers the APP, and comprises information such as gender, age, income, native place, living area and the like.

The user social attribute information is obtained by analyzing user behavior information received in real time, and comprises information such as the activity of logging in an APP, preference categories, purchasing power and order amount. The user behavior information is behaviors of browsing, clicking, collecting, praising, buying all and the like generated by the user logging in the APP.

It should be noted that, because the behavior information generated by the user in the APP and/or the filled personal information change with time, the background server may dynamically update the user portrait once at regular intervals, so that the user portrait can better reflect the current characteristics of the user.

In an optional embodiment, in the screening process for the candidate users, for each target user, the push probability may be predicted according to the user figure of the target user, and the candidate users may be screened from the target users according to the push probability.

The push probability represents the information push value degree of the target user, and the larger the push probability is, the higher the information push value is.

Based on this, when candidate users are screened according to the push probability, the target users with the push probability larger than the second threshold value can be selected from all the target users as the candidate users.

Optionally, in the prediction process of the push probability, a first specified type of user features may be extracted based on the user portrait of the target user, and the user features are input into the trained first classification model, so that the first classification model predicts the push probability of the target user according to the user features and outputs the push probability, thereby converting the similarity problem between predicted users into a two-classification problem, and expanding the audience coverage of information.

The first specific type of user features may be a set of multiple features, or may be a single feature that can reflect user characteristics of the user in terms of purchasing power, activity, gender, age, and the like. In order to enable the first classification model to output probability values instead of classification results, a sigmoid function is added to the output layer of the first classification model to convert the classification results into probability output.

It is worth noting that before the first classification model is applied, the first classification model needs to be trained in advance, for the training process of the first classification model, firstly, user features of a first specified type are extracted as positive sample features based on user figures of users who have made orders, and the user features of the first specified type are extracted as negative sample features based on user figures of users who have not made orders and have liveness higher than a first threshold, and then the constructed first classification model is trained by using the positive sample features and the negative sample features until convergence.

The ordered users as the positive samples are users who have generated purchasing behaviors, and the users as the negative samples are users who log in the APP frequently and have not generated purchasing behaviors.

Further, before training, it is also necessary to set a positive sample label for the positive sample feature, for example, setting "1" to indicate a positive sample, and setting a negative sample label for the negative sample feature, for example, setting "0" to indicate a negative sample.

It can be understood by those skilled in the art that the structure of the first classification model for predicting the push probability is not specifically limited, and may be implemented by using any existing classification model structure, for example, any one of a random forest RF, a gradient boost decision tree GBDT, and an XGBoost classification model may be used.

Step 102: and acquiring the push value and the intention probability of each candidate user aiming at the specified category, wherein the push value and the intention probability are both obtained by analyzing the user portrait.

The specified category is a category to which the information to be pushed belongs. In the medical and cosmetic field, categories may include double eyelids, eyebrow tattoos, eyebrow lifts, and the like.

The push value of the user and the intention probability aiming at the specified category are obtained by analyzing the user portrait, so that the background server can analyze the push value of the user and the intention probability aiming at the specified category in advance and dynamically update the push value and the intention probability once at regular intervals. Therefore, when the device is used, the device can directly acquire the push value of the candidate user and the intention probability aiming at the specified category.

As will be appreciated by those skilled in the art, since user portraits may change over time, in order to ensure real-time push value and intent probability, current user portraits may be re-analyzed during use.

In an optional embodiment, the push value of the user is related to purchasing ability, so that the push value can be predicted directly according to the characteristic of the order placing amount in the user image, that is, for each candidate user, if the candidate user generates an order placing behavior, the average transaction amount of the candidate user is obtained as the push value, and if the candidate user does not generate the order placing behavior, the second preset value is used as the push value, and the prediction mode is simple and easy to operate.

Wherein, the second preset value is a default value set for a user who does not generate ordering behavior.

It should be noted that, in the above-mentioned push value prediction method, only one characteristic of the user is considered, and in order to improve the prediction accuracy of the push value, it is also necessary to consider other characteristics of the user for prediction, and a specific prediction process may be referred to in the following description of the embodiments, which is not detailed herein.

Further, for the prediction process of the intention probability of the specified category, reference may be made to the following description of the embodiments, which will not be detailed herein.

Step 103: and screening users for pushing information under the appointed category from the candidate users according to the pushing value and the intention probability.

The users screened from the candidate users belong to the extended crowd aiming at the specified category, and the click rate and the conversion rate of the information can be improved by pushing the information under the specified category to the extended crowd.

In an alternative embodiment, the candidate users with the pushing value higher than the third threshold and the intention probability higher than the fourth threshold may be selected as the users for pushing the information under the specified category.

So far, the screening process shown in fig. 1 is completed, after candidate users are screened from target users, the candidate users are further screened according to the pushing value and the intention probability by obtaining the pushing value and the intention probability of the candidate users for the designated class, so that the pushing precision of the crowd is improved, and the coverage and the effectiveness of pushing are further improved. And the intention probability of the user aims at the category to which the information to be pushed belongs, so that the targeted final user is stronger, and the pushing precision of the crowd can be further improved.

The second embodiment:

fig. 2 is a schematic diagram illustrating a process of predicting a push value of a candidate user according to the embodiment shown in fig. 1, where based on the embodiment shown in fig. 1, as shown in fig. 2, a process of predicting a push value of a candidate user includes the following steps:

step 201: a second specified type of user feature is extracted based on the user representation of the candidate user.

In this embodiment, the second specified type is a subset of the first specified type, and the user features of the second specified type have a higher degree of importance than the user features of the other types of the first specified type except the second specified type.

Optionally, before step 201 is executed, the importance ranking may be performed on the user features of the first specified type, and then a certain number of user features ranked at the top are selected as the optimal feature subset, that is, the user features of the second specified type.

Those skilled in the art will understand that the process of ranking importance of features may be implemented by using a related technology, which is not specifically limited in this application, for example, the importance ranking may be performed by using an XGBoost embedded feature selection method.

Step 202: and respectively inputting the user characteristics of the second specified type into different regression models so as to predict the sub-push values of the candidate users according to the user characteristics by each regression model.

In this embodiment, in order to improve the accuracy of the push value, regression predictions are performed by using a plurality of different regression models, and the push values of the candidate users are obtained by integrating the prediction results of the plurality of different regression models.

Before step 202 is executed, for the training process of each regression model, a user feature of a second specified type may be extracted as a positive sample feature based on the user portrait of the ordered user, and an average transaction amount of the ordered user is obtained as a sample label corresponding to the positive sample feature, while a user feature of a second specified type is extracted as a negative sample feature based on the user portrait of the unordered user, and a first preset value is used as a sample label corresponding to the negative sample feature, and then each regression model is trained separately by using the positive sample feature and its corresponding sample label, and the negative sample feature and its corresponding sample label until convergence.

For the non-ordering user, since the ordering transaction amount is 0, a default value (i.e. a first preset value) can be set as the sample label for the non-ordering user.

Alternatively, three different regression models may be used for regression prediction, taking into account the prediction complexity and the device performance.

Specifically, the three different regression models may be three basis learners of random forest RF, gradient boosting decision trees GBDT, and XGBoost.

It should be noted that, in order to improve the accuracy and stability of the model and avoid the occurrence of overfitting of the model, after the first training is completed, each regression model may be trained again by predicting the training samples and using the prediction result as the training sample for the second training.

After each regression model is trained by using the positive sample characteristics and the corresponding sample labels thereof, and the negative sample characteristics and the corresponding sample labels thereof, the negative sample characteristics of the non-lower single user are respectively input into each trained regression model, the sub-push value of the non-lower single user is predicted by each trained regression model according to the negative sample characteristics, then the push value of the non-lower single user is determined according to the sub-push value output by each regression model, the sample labels corresponding to the negative sample characteristics are updated by using the push value, and each trained regression model is retrained by using the negative sample characteristics and the updated sample labels corresponding to the negative sample characteristics until convergence.

Step 203: and determining the push value of the candidate user according to the sub-push values output by each regression model.

Optionally, the individual sub-push values may be weighted and summed to obtain the push value of the candidate user.

By this point, the prediction process shown in fig. 2 is completed, and the sub-push values obtained by using the multiple regression models to predict respectively are used to determine the push values of the candidate users, so that the accuracy of the push values can be improved, and the user values can be better measured. And the user characteristics of the second specified type input into each regression model comprise different dimensional characteristics of the user, and the accuracy can be further improved by comprehensively considering the prediction push value of the different dimensional characteristics of the user.

Example three:

fig. 3 is a schematic diagram illustrating a process of predicting the intention probability of a candidate user for a specified category according to the embodiment shown in fig. 1, where based on the embodiments shown in fig. 1 to fig. 2, as shown in fig. 3, the process of predicting the intention probability for the specified category includes the following steps:

step 301: a first specified type of user feature is extracted based on a user representation of the candidate user.

For the extraction process of the user features of the first specified type, reference may be made to the related description of step 101, which is not described herein again.

Step 302: and inputting the user characteristics of the first specified type into a second classification model corresponding to the specified category so as to predict the intention probability of the candidate user for the specified category according to the user characteristics by the second classification model.

In the embodiment, in order to ensure the accuracy of the probability of intention for a specific category, a special second classification model is used for prediction for each category.

It should be noted that the second classification model may have the same structure as the first classification model, and although both the first classification model and the second classification model are used for predicting the probability, they are trained by using different training samples, so that the prediction pertinence of the two models is different.

Optionally, the training process for the second classification model includes: and acquiring the ordered user under the appointed category, extracting a user characteristic of a first appointed type as a positive sample characteristic based on the acquired user portrait of the ordered user, extracting a user characteristic of a first appointed type as a negative sample characteristic based on the user portrait of the unordered user, and training the constructed second classification model by using the positive sample characteristic and the negative sample characteristic until convergence.

Before training, it is also necessary to set a positive sample label for the positive sample feature, for example, setting to "1" indicates a positive sample, and to set a negative sample label for the negative sample feature, for example, setting to "0" indicates a negative sample.

Therefore, the positive sample source for training the first classification model is from the ordered users under all categories, the positive sample source for training the second classification model is from the ordered users under the designated category, the negative sample source for training the first classification model is from more active unordered users, and the negative sample source for training the second classification model is from only unordered users.

So far, the above-mentioned prediction process shown in fig. 3 is completed, a special second classification model is correspondingly set for each category, and the second classification model corresponding to the specified category is used to predict the intention probability of the candidate user for the specified category.

Corresponding to the embodiment of the user screening method, the invention also provides an embodiment of a user screening device.

Fig. 4 is a schematic structural diagram of a user screening apparatus according to an exemplary embodiment of the present invention, the apparatus is configured to execute the user screening method provided in any of the above embodiments, and as shown in fig. 4, the user screening apparatus includes:

the first screening module 410 is used for acquiring user figures of all target users and screening candidate users from all target users according to the user figures;

an obtaining module 420, configured to obtain a push value and an intention probability for a specified category of each candidate user; the push value and the intention probability are obtained by analyzing the user portrait;

and a second filtering module 430, configured to filter, according to the push value and the intention probability, users who are used to push information under the specified category from the candidate users.

The implementation process of the functions and actions of each unit in the above device is specifically described in the implementation process of the corresponding step in the above method, and is not described herein again.

For the device embodiments, since they substantially correspond to the method embodiments, reference may be made to the partial description of the method embodiments for relevant points. The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of the invention. One of ordinary skill in the art can understand and implement it without inventive effort.

The embodiment of the invention also provides electronic equipment corresponding to the user screening method provided by the embodiment, so as to execute the user screening method.

Fig. 5 is a hardware block diagram of an electronic device according to an exemplary embodiment of the present invention, the electronic device including: a communication interface 601, a processor 602, a memory 603, and a bus 604; the communication interface 601, the processor 602 and the memory 603 communicate with each other via a bus 604. The processor 602 may execute the user screening method described above by reading and executing machine executable instructions corresponding to the control logic of the user screening method in the memory 603, and the details of the method are described in the above embodiments, which will not be described herein again.

The memory 603 referred to in this disclosure may be any electronic, magnetic, optical, or other physical storage device that can contain stored information, such as executable instructions, data, and so forth. Specifically, the Memory 603 may be a RAM (Random Access Memory), a flash Memory, a storage drive (e.g., a hard disk drive), any type of storage disk (e.g., an optical disk, a DVD, etc.), or similar storage medium, or a combination thereof. The communication connection between the network element of the system and at least one other network element is realized through at least one communication interface 601 (which may be wired or wireless), and the internet, a wide area network, a local network, a metropolitan area network, and the like can be used.

Bus 604 may be an ISA bus, PCI bus, EISA bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 603 is used for storing a program, and the processor 602 executes the program after receiving the execution instruction.

The processor 602 may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware or instructions in the form of software in the processor 602. The Processor 602 may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; but may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components. The various methods, steps, and logic blocks disclosed in the embodiments of the present application may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present application may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor.

The electronic equipment provided by the embodiment of the application and the user screening method provided by the embodiment of the application are based on the same inventive concept, and have the same beneficial effects as the method adopted, operated or realized by the electronic equipment.

Referring to fig. 6, the illustrated computer-readable storage medium is an optical disc 30, on which a computer program (i.e., a program product) is stored, and when the computer program is executed by a processor, the computer program may execute the user filtering method provided in any of the foregoing embodiments.

It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, a phase change memory (PRAM), a Static Random Access Memory (SRAM), a Dynamic Random Access Memory (DRAM), other types of Random Access Memories (RAM), a Read Only Memory (ROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), a flash memory, or other optical and magnetic storage media, which are not described in detail herein.

The computer-readable storage medium provided by the above-mentioned embodiment of the present application and the user screening method provided by the embodiment of the present application have the same beneficial effects as the method adopted, operated or implemented by the application program stored in the computer-readable storage medium.

Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.

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 a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (15)

1. A method for user screening, the method comprising:

acquiring a user portrait of each target user, and screening candidate users from each target user according to the user portrait;

acquiring the push value of each candidate user and the intention probability aiming at the specified category; the pushing value and the intention probability are obtained by analyzing the user portrait;

and screening users for pushing the information under the appointed category from the candidate users according to the pushing value and the intention probability.

2. The method of claim 1, wherein the filtering candidate users from respective target users according to user profile comprises:

for each target user, predicting a push probability according to a user portrait of the target user;

and screening candidate users from all target users according to the push probability.

3. The method of claim 2, wherein predicting a push probability from a user profile of the target user comprises:

extracting a first specified type of user feature based on a user representation of the target user;

inputting the user characteristics into a trained first classification model, and predicting and outputting the pushing probability of a target user according to the user characteristics by the first classification model.

4. The method of claim 3, wherein the training process for the first classification model comprises:

extracting user features of a first specified type as positive sample features based on the user portrait of the ordered user;

extracting a first specified type of user features as negative sample features based on a user portrait of a user who has not placed a order and has a liveness higher than a first threshold;

and training the constructed first classification model by using the positive sample characteristics and the negative sample characteristics until convergence.

5. The method of claim 1, wherein the screening candidate users from the target users according to the push probability comprises:

and selecting the target users with the push probability larger than a second threshold value from all the target users as candidate users.

6. The method of claim 1, wherein obtaining the push value of each candidate user comprises:

for each candidate user, extracting a second specified type of user feature based on a user representation of the candidate user;

inputting the user characteristics of the second specified type into different regression models respectively so that each regression model can predict the sub-push value of the candidate user according to the user characteristics;

and determining the push value of the candidate user according to the sub-push values output by each regression model.

7. The method of claim 6, wherein the training process for each regression model comprises:

extracting a user characteristic of a second specified type as a positive sample characteristic based on the user portrait of the ordered user, and acquiring the average transaction amount of the ordered user as a sample label corresponding to the positive sample characteristic;

extracting user features of a second specified type as negative sample features based on the user portrait of the user who does not give the order, and taking a first preset value as a sample label corresponding to the negative sample features;

and respectively training each regression model by utilizing the positive sample characteristics and the corresponding sample labels thereof and the negative sample characteristics and the corresponding sample labels thereof until convergence.

8. The method of claim 7, wherein after each regression model is separately trained using positive sample features and their corresponding sample labels, negative sample features and their corresponding sample labels, the method further comprises:

respectively inputting the negative sample characteristics of the users who do not order into each trained regression model, and predicting the sub-push value of the users who do not order according to the negative sample characteristics by each trained regression model;

determining the push value of the users who do not leave the order according to the sub-push value output by each regression model, and updating the sample label corresponding to the negative sample feature by using the push value;

and (4) retraining each trained regression model respectively by using the negative sample characteristics and the corresponding updated sample labels until convergence.

9. The method of claim 1, wherein obtaining the push value of each candidate user comprises:

for each candidate user, if the candidate user generates ordering behavior, acquiring the average transaction amount of the candidate user as a push value;

and if the candidate user does not generate ordering behavior, taking the second preset value as a pushing value.

10. The method of claim 1, wherein obtaining an intention probability of each candidate user for a specified category comprises:

for each candidate user, extracting a first specified type of user features based on a user portrait of the candidate user;

and inputting the user characteristics of the first specified type into a second classification model corresponding to the specified category so as to predict the intention probability of the candidate user for the specified category according to the user characteristics by the second classification model.

11. The method of claim 10, wherein the training process for the second classification model comprises:

acquiring the ordered user under the specified category, and extracting a user feature of a first specified type as a positive sample feature based on the acquired user portrait of the ordered user;

extracting user features of a first specified type as negative sample features based on user figures of users who have not made orders;

and training the constructed second classification model by using the positive sample characteristics and the negative sample characteristics until convergence.

12. The method according to claim 1, wherein the screening users from candidate users for pushing information under the specified category according to the pushing value and the intention probability comprises:

and selecting candidate users with push values higher than a third threshold value and intention probabilities higher than a fourth threshold value from the candidate users as the users for pushing the information under the specified category.

13. A user screening apparatus, the apparatus comprising:

the first screening module is used for acquiring the user portrait of each target user and screening candidate users from each target user according to the user portrait;

the acquisition module is used for acquiring the push value of each candidate user and the intention probability aiming at the specified category; the push value and the intention probability are obtained by analyzing the user portrait;

and the second screening module is used for screening the users for pushing the information under the specified category from the candidate users according to the pushing value and the intention probability.

14. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method according to any of claims 1-12 are implemented when the processor executes the program.

15. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 12.

Images