CN112348592A

CN112348592A - Advertisement recommendation method and device, electronic equipment and medium

Info

Publication number: CN112348592A
Application number: CN202011330780.8A
Authority: CN
Inventors: 顾立瑞
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-11-24
Filing date: 2020-11-24
Publication date: 2021-02-09

Abstract

The application provides an advertisement recommendation method, an advertisement recommendation device, electronic equipment and a medium. The method comprises the following steps: determining the matching probability of each advertisement to be recommended aiming at each user based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group; acquiring a conversion success reward value corresponding to each advertisement to be recommended; calculating the total reward value of each advertisement to be recommended aiming at the user group based on the matching probability and the conversion success reward value; and selecting target advertisements from the advertisements to be recommended to recommend the advertisements to the user group based on the total reward value. According to the technical scheme, when advertisement recommendation is carried out, the reward value and the income of the user group are considered, and then more accurate advertisements can be selected to be recommended to the user group, so that the accuracy of advertisement recommendation is improved, and the conversion rate of the advertisements can be improved.

Description

Advertisement recommendation method and device, electronic equipment and medium

Technical Field

The present application relates to the field of computer and communications technologies, and in particular, to an advertisement recommendation method, an apparatus, an electronic device, and a medium.

Background

With the gradual development of social informatization, advertisements gradually become a way to lead user demands, and how to successfully attract customers becomes an inevitable problem for each advertiser.

In the prior art, an advertiser usually puts advertisements in a position with a large browsing amount, and attracts customers by increasing the exposure times of the advertisements, and the advertisement recommendation method is poor in recommendation accuracy and low in conversion rate.

Disclosure of Invention

The application aims to provide an advertisement recommendation method, an advertisement recommendation device, electronic equipment and a medium, which can improve the accuracy of advertisement recommendation to a certain extent and can improve the conversion rate of advertisements to a certain extent.

According to an aspect of an embodiment of the present application, there is provided an advertisement recommendation method, including: determining the matching probability of each advertisement to be recommended aiming at each user based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group; acquiring a conversion success reward value corresponding to each advertisement to be recommended; calculating the total reward value of each advertisement to be recommended aiming at the user group based on the matching probability and the conversion success reward value; and selecting target advertisements from the advertisements to be recommended to recommend to the user group based on the overall reward value.

According to an aspect of an embodiment of the present application, there is provided an advertisement recommendation apparatus including: the matching module is configured to determine the matching probability of each advertisement to be recommended aiming at each user based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group; the acquisition module is configured to acquire the conversion success reward value corresponding to each advertisement to be recommended; the calculation module is used for calculating the total reward value of each advertisement to be recommended aiming at the user group based on the matching probability and the conversion success reward value; and the recommending module is configured to select a target advertisement from the advertisements to be recommended and recommend the target advertisement to the user group based on the overall reward value.

In one embodiment of the present application, the calculation module is configured to: calculating the product of the conversion success reward value corresponding to each advertisement to be recommended and the matching probability of each advertisement to be recommended aiming at each user to obtain the reward value of each advertisement to be recommended aiming at each user; and summing the reward values of the advertisements to be recommended aiming at the users to obtain the total reward value of the advertisements to be recommended aiming at the user group.

In one embodiment of the present application, the recommendation module is configured to: selecting a set number of advertisements to be recommended from the advertisements to be recommended as the target advertisements according to the sequence of the overall reward value from large to small; recommending the targeted advertisement to the user group.

In one embodiment of the present application, the matching module is configured to: calculating the characteristic distance between the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended; and mapping the characteristic distance through an activation function to obtain the matching probability of each advertisement to be recommended aiming at each user.

In one embodiment of the present application, the matching module is further configured to: before calculating the characteristic distance between the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended, mapping the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended to the same semantic space.

In one embodiment of the present application, the matching module is configured to: acquiring the size of a user characteristic graph corresponding to the user characteristic of each user and the size of an advertisement characteristic graph corresponding to the advertisement characteristic of each advertisement to be recommended; and performing matrix transformation on the convolution layer with the user characteristic input convolution kernel size of each user as the user characteristic graph size, and performing matrix transformation on the convolution layer with the advertisement characteristic input convolution kernel size of each advertisement to be recommended as the advertisement characteristic graph size so as to convert the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended into the same semantic space.

In one embodiment of the present application, each of the users has a plurality of user characteristics, and the plurality of user characteristics includes: the system comprises a sequence type user characteristic used for describing the relation between user behavior and time variation, a continuous type user characteristic continuously changing along with the time variation, and a discrete type user characteristic not continuously changing along with the time variation; each advertisement to be recommended has a plurality of advertisement characteristics, wherein the plurality of advertisement characteristics comprise: a continuous advertisement feature that continuously changes with time, and a discrete advertisement feature that does not continuously change with time; the matching module is further configured to: and splicing the multiple user characteristics of each user and the multiple advertisement characteristics of each advertisement to be recommended before mapping the user characteristics of each user and the advertisement characteristics of each advertisement to be recommended to the same semantic space.

In one embodiment of the present application, the matching module is configured to: inputting the advertisement characteristics of each advertisement to be recommended to a first characteristic processing branch of a neural network model, and inputting the user characteristics of each user to a second characteristic processing branch of the neural network model, wherein the first characteristic processing branch and the second characteristic processing branch are connected to a comprehensive processing unit; and acquiring the matching probability of each advertisement to be recommended, which is output by the comprehensive processing unit, aiming at each user.

In one embodiment of the present application, the first feature processing branch comprises: the first characteristic cross module and the first characteristic splicing module are connected with the output end of the first characteristic cross module; the first feature crossing module is used for performing feature crossing processing on discrete advertisement features in the advertisement features, and the advertisement splicing module is used for splicing continuous advertisement features in the advertisement features and the advertisement crossing features output by the first feature crossing module; the second feature processing branch comprises: the time sequence feature extraction module is connected with the output end of the second feature crossing module; the second feature crossing module is configured to perform feature crossing processing on discrete user features in the user features, the time sequence feature extraction module is configured to perform self-attention calculation on sequence-type user features in the user features, and the user splicing module is configured to splice continuous user features in the user features, the user crossing features output by the second feature crossing module, and the self-attention features output by the time sequence feature extraction module.

In one embodiment of the present application, the neural network model is pre-trained by: acquiring a user characteristic sample set and an advertisement characteristic sample set, and matching probabilities between each user characteristic sample in the user characteristic sample set and each advertisement characteristic sample in the advertisement characteristic sample set; inputting the user characteristic samples into the first characteristic processing branch, inputting the advertisement characteristic samples into the second characteristic processing branch, and obtaining the prediction probability between each user characteristic sample in the user characteristic sample set and each advertisement characteristic sample in the advertisement characteristic sample set output by the neural network model; comparing the predicted probability with the matching probability, and if not, adjusting the neural network model so that the predicted probability is consistent with the matching probability.

In one embodiment of the present application, the advertisement recommendation apparatus further includes: a synthesis module configured to: taking the user characteristic sample and the advertisement characteristic sample with the matching probability reaching the set probability as positive samples; taking the user characteristic sample and the advertisement characteristic sample with the matching probability not reaching the set probability as negative samples; if the ratio of the number of the positive samples to the number of the negative samples does not reach a threshold value, calculating a first characteristic distance between each positive sample and any one of the positive samples, and randomly selecting the positive sample with the characteristic distance smaller than the first characteristic distance as a synthesized positive sample; until the number of negative samples reaches the threshold in a ratio of the number of positive samples to the number of composite positive samples.

According to an aspect of embodiments of the present application, there is provided a computer-readable program medium storing computer program instructions which, when executed by a computer, cause the computer to perform the method of any one of the above.

According to an aspect of an embodiment of the present application, there is provided an electronic apparatus including: a processor; a memory having computer readable instructions stored thereon which, when executed by the processor, implement the method of any of the above.

According to an aspect of embodiments herein, there is provided a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the advertisement recommendation method provided in the various alternative embodiments described above.

The technical scheme provided by the embodiment of the application can have the following beneficial effects:

in the technical scheme provided by some embodiments of the application, based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group, the matching probability of each advertisement to be recommended for each user is determined, the conversion success reward value corresponding to each advertisement to be recommended is obtained, based on the matching probability and the conversion success reward value, the overall reward value of each advertisement to be recommended for the user group is calculated, and based on the overall reward value, a target advertisement is selected from the advertisements to be recommended and recommended to the user group. Therefore, according to the technical scheme of the embodiment of the application, the proper advertisements are recommended to the user groups, the reward value and the income of the user groups can be considered, and then more accurate advertisements can be selected to be recommended to the user groups, so that the accuracy of advertisement recommendation is improved, and the conversion rate of the advertisements can be improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and together with the description, serve to explain the principles of the application.

FIG. 1 shows a schematic diagram of an exemplary system architecture to which aspects of embodiments of the present application may be applied;

FIG. 2 schematically illustrates a flow diagram of a method of advertisement recommendation according to an embodiment of the present application;

FIG. 3 schematically shows a structural diagram of a neural network model according to an embodiment of the present application;

FIG. 4 schematically illustrates a structural schematic of a first feature crossover module according to one embodiment of the present application;

FIG. 5 schematically shows a structural diagram of a temporal feature extraction module according to one embodiment of the present application;

FIG. 6 schematically illustrates a flow diagram of a method of advertisement recommendation according to an embodiment of the present application;

FIG. 7 schematically illustrates a block diagram of an advertisement recommendation device according to an embodiment of the present application;

FIG. 8 is a hardware diagram illustrating an electronic device according to an example embodiment.

Detailed Description

Example embodiments will now be described more fully with reference to the accompanying drawings. Example embodiments may, however, be embodied in many different forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art.

Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the application. One skilled in the relevant art will recognize, however, that the subject matter of the present application can be practiced without one or more of the specific details, or with other methods, components, devices, steps, and so forth. In other instances, well-known methods, devices, implementations, or operations have not been shown or described in detail to avoid obscuring aspects of the application.

The block diagrams shown in the figures are functional entities only and do not necessarily correspond to physically separate entities. I.e. these functional entities may be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different networks and/or processor means and/or microcontroller means.

The flow charts shown in the drawings are merely illustrative and do not necessarily include all of the contents and operations/steps, nor do they necessarily have to be performed in the order described. For example, some operations/steps may be decomposed, and some operations/steps may be combined or partially combined, so that the actual execution sequence may be changed according to the actual situation.

Cloud technology refers to a hosting technology for unifying serial resources such as hardware, software, network and the like in a wide area network or a local area network to realize calculation, storage, processing and sharing of data.

Cloud technology (Cloud technology) is based on a general term of network technology, information technology, integration technology, management platform technology, application technology and the like applied in a Cloud computing business model, can form a resource pool, is used as required, and is flexible and convenient. Cloud computing technology will become an important support. Background services of the technical network system require a large amount of computing and storage resources, such as video websites, picture-like websites and more web portals. With the high development and application of the internet industry, each article may have its own identification mark and needs to be transmitted to a background system for logic processing, data in different levels are processed separately, and various industrial data need strong system background support and can only be realized through cloud computing.

Big data (Big data) refers to a data set which cannot be captured, managed and processed by a conventional software tool within a certain time range, and is a massive, high-growth-rate and diversified information asset which can have stronger decision-making power, insight discovery power and flow optimization capability only by a new processing mode. With the advent of the cloud era, big data has attracted more and more attention, and the big data needs special technology to effectively process a large amount of data within a tolerance elapsed time. The method is suitable for the technology of big data, and comprises a large-scale parallel processing database, data mining, a distributed file system, a distributed database, a cloud computing platform, the Internet and an extensible storage system.

The big data can be applied to content recommendation, and due to the strong data processing capacity, the efficiency and the accuracy of content recommendation can be remarkably improved, the advertisement recommendation scheme related in the embodiment of the application can guarantee the accuracy of advertisement recommendation based on the cloud technology and the big data, and the detailed description is as follows:

fig. 1 shows a schematic diagram of an exemplary system architecture 100 to which the technical solutions of the embodiments of the present application can be applied.

As shown in fig. 1, the system architecture 100 may include a client 101 (the sending client may be one or more of a smartphone, a tablet, a laptop, a desktop computer), a network 102, and a server 103. Network 102 is the medium used to provide communication links between sending clients 101 and servers. Network 102 may include various connection types, such as wired communication links, wireless communication links, and so forth.

It should be understood that the number of clients 101, networks 102, and servers 103 in fig. 1 is merely illustrative. There may be any number of clients 101, networks 102, and servers 103, as desired for implementation. For example, the server 103 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, a middleware service, a domain name service, a security service, a CDN (Content Delivery Network), a big data and artificial intelligence platform, and the like. The terminal may be, but is not limited to, a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and the like. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the application is not limited herein.

In an embodiment of the application, the server 103 may obtain user characteristics of each user in the user group through the client 101 where each user in the user group is located, determine matching probabilities of each advertisement to be recommended for each user based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group, obtain a conversion success reward value corresponding to each advertisement to be recommended, calculate a total reward value of each advertisement to be recommended for the user group based on the matching probabilities and the conversion success reward values, select a target advertisement from the advertisements to be recommended to the client 101 where the user in the user group is located based on the total reward value, recommend a suitable advertisement to the user group while considering the reward value revenue of the user group, can attract the conversion of the users in the user group, and further can select a more accurate advertisement to recommend to the user group, the accuracy of advertisement recommendation is improved, and the conversion rate of the advertisement can be improved.

It should be noted that the advertisement recommendation method provided in the embodiment of the present application is generally executed by the server 103, and accordingly, the advertisement recommendation device is generally disposed in the server 103. However, in other embodiments of the present application, the client 101 may also have a similar function as the server 103, so as to execute the advertisement recommendation method provided by the embodiments of the present application.

The implementation details of the technical solution of the embodiment of the present application are set forth in detail below:

fig. 2 schematically shows a flowchart of an advertisement recommendation method according to an embodiment of the present application, where an execution subject of the advertisement recommendation method may be a server, such as the server 103 shown in fig. 1.

Referring to fig. 2, the advertisement recommendation method at least includes steps S210 to S240, which are described in detail as follows:

in step S210, a matching probability of each advertisement to be recommended for each user is determined based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group.

In one embodiment of the present application, the advertisement characteristics may include advertisement length, advertisement type, advertisement audience type, average browsing duration, click number, exposure number, and the like, and the dimension of the advertisement characteristics may be more than ten thousand, so that the obtained advertisement characteristics are more comprehensive.

In one embodiment of the present application, each advertisement to be recommended may have a plurality of advertisement features, and the plurality of advertisement features may include: a continuous advertising feature that continuously changes over time, and a discrete advertising feature that does not continuously change over time. The continuous type advertisement characteristics may include an average browsing duration, a number of clicks, a number of exposures, and the like. The discrete ad features may include ad length, ad type, and ad audience type, among others.

In an embodiment of the present application, a user group may include one or more users, one user group may correspond to one chat interface, and each user in the user group may send a message to the chat interface corresponding to the user group, or may receive a message sent to the chat interface by another user in the same user group.

In one embodiment of the present application, the user characteristics may include user basic characteristics and user interest characteristics. The user basic characteristics may include a user name, gender, age, location, education level, a terminal model, and the like. The user interest characteristics can comprise commercial interests, marketing interests, application program interests, media interests, business district category interests and the like of the user, and the dimension of the user characteristics can be more than ten thousand, so that the acquired advertisement characteristics are more comprehensive.

In one embodiment of the present application, each user may have a plurality of user characteristics, which may include: the system comprises a sequence type user characteristic used for describing the relation between user behaviors and time variation, a continuous type user characteristic continuously changing along with the time variation, and a discrete type user characteristic continuously changing along with the time variation. The sequential user features may include: data representing the time-varying relationship of the user's purchasing characteristics for different kinds of items, data representing the time-varying relationship of the user's click-to-browse characteristics for different kinds of items, data representing the time-varying relationship of the user's interest in the application, etc., for example, a sequential type user characteristic may include the kind of items that the user has clicked to browse daily or weekly in the past month. The continuous user characteristics may include the age of the user, the length of time the user views advertisements daily, and the like.

In an embodiment of the application, a feature distance between a user feature of each user and an advertisement feature of each advertisement to be recommended may be calculated, and the feature distance is mapped through an activation function, so as to obtain a matching probability of each advertisement to be recommended for each user. The activation function may map the feature distance to a range from zero to one, so as to obtain a matching probability corresponding to the feature distance, which is used as a matching probability of each advertisement to be recommended for each user.

In one embodiment of the present application, the characteristic distance may be an euclidean distance, a manhattan distance, a chebyshev distance, a minkowski distance, a normalized euclidean distance, a mahalanobis distance, an included cosine, a hamming distance, a jaccard distance, a correlation distance, or an information entropy.

In one embodiment of the present application, the activation function may be a Sigmoid function, a hyperbolic tangent (Tanh) function, or a Linear rectification function (ReLU).

In an embodiment of the application, before calculating the feature distance between the user feature of each user and the advertisement feature of each advertisement to be recommended, the user feature of each user and the advertisement feature of each advertisement to be recommended may be mapped to the same semantic space, so as to more conveniently calculate the matching probability between the user feature of each user and the advertisement feature of each advertisement to be recommended.

In an embodiment of the application, the size of a user feature map corresponding to the user feature of each user and the size of an advertisement feature map corresponding to the advertisement feature of each advertisement to be recommended may be obtained, matrix transformation is performed on a convolution layer in which the size of a user feature input convolution kernel of each user is the size of the user feature map, matrix transformation is performed on a convolution layer in which the size of an advertisement feature input convolution kernel of each advertisement to be recommended is the size of the advertisement feature map, and the user feature of each user and the advertisement feature of each advertisement to be recommended are transformed into the same semantic space.

In an embodiment of the application, before mapping the user characteristics of each user and the advertisement characteristics of each advertisement to be recommended to the same semantic space, the multiple user characteristics of each user can be spliced, and the multiple advertisement characteristics of each advertisement to be recommended can be spliced, so that the matching probability between the user characteristics of each user and the advertisement characteristics of each advertisement to be recommended can be calculated more conveniently.

In an embodiment of the present application, a self-attention calculation may be performed on every two sequence-type user features of each user to obtain a self-attention feature, and the self-attention feature may embody time sequence information in the sequence-type user features. The method can perform feature cross processing on every two discrete user features of each user to obtain user cross features, and the user cross features can embody the relationship between every two discrete user features of each user. The continuous user characteristics, the self-attention characteristics and the user cross characteristics of each user can be spliced, so that the obtained splicing information can reflect deeper information contained in the user characteristics.

In an embodiment of the present application, after the same type of user features of each user are spliced, the user features of each user are spliced with different types of user features of each user, for example, the continuous type user features, the self-attention features, and the user cross features of each user may be respectively spliced, and then the spliced continuous type user features, the spliced self-attention features, and the spliced user cross features of each user may be spliced.

In an embodiment of the application, feature crossing processing may be performed on every two discrete advertisement features of each advertisement to be recommended to obtain an advertisement crossing feature, and the advertisement crossing feature may embody a relationship between every two discrete advertisement features of each advertisement. And splicing the continuous advertisement characteristics and the advertisement cross characteristics of each advertisement to be recommended, so that the obtained splicing information can reflect deeper information contained in the advertisement characteristics.

In an embodiment of the present application, after splicing the same kind of advertisement features of each advertisement to be recommended, the advertisement features may be spliced with different kinds of advertisement features of each advertisement to be recommended, for example, the continuous advertisement features, the self-attention features, and the advertisement cross features of each advertisement to be recommended may be spliced, and then the continuous advertisement features, the spliced self-attention features, and the spliced advertisement cross features of each advertisement to be recommended may be spliced.

With continued reference to fig. 2, in step S220, a conversion success reward value corresponding to each advertisement to be recommended is obtained.

In an embodiment of the present application, the conversion success reward value corresponding to each advertisement to be recommended is set by the advertiser, and is a reward value provided by the advertiser when each advertisement to be recommended is converted successfully once, and the reward value may be a virtual reward or an entity reward, for example, the reward value may be a fund in an account.

In one embodiment of the present application, the conversion success reward value may be obtained by the converted user to attract the user to be converted.

In one embodiment of the present application, the conversion success reward value may be divided by members in the user group, and may be a value obtained by the converted users obtaining part of the conversion success reward value, and the remaining conversion success reward values may be distributed according to the number of times that the users are converted by the advertisements in the user group, or may be distributed according to the conversion success reward values that the users have obtained in the user group, so as to attract the users to be converted in the user group.

In an embodiment of the application, the conversion success reward value may be obtained by an administrator of the user group to attract the administrator of the user group to cooperate with advertisement recommendation, the administrator of the user group generally has a certain influence on other people in the user group, and the conversion rate can be improved after the advertisement to be recommended is cooperated by the administrator.

In step S230, based on the matching probability and the conversion success reward value, an overall reward value of each advertisement to be recommended for the user group is calculated.

In an embodiment of the application, a product between a conversion success reward value corresponding to each advertisement to be recommended and a matching probability of each advertisement to be recommended for each user may be calculated to obtain a reward value of each advertisement to be recommended for each user, and the reward values of each advertisement to be recommended for each user are summed to obtain an overall reward value of each advertisement to be recommended for a user group.

In one embodiment of the present application, a weighted sum of the matching probability and the conversion success reward value may be calculated to obtain an overall reward value of each advertisement to be recommended for the user group.

In step S240, a target advertisement recommendation is selected from the advertisements to be recommended and is recommended to the user group based on the overall reward value.

In an embodiment of the application, a set number of advertisements to be recommended may be selected from the advertisements to be recommended as target advertisements in the order from large to small of the overall reward value, and the target advertisements are recommended to the user group.

In an embodiment of the application, the to-be-recommended list may be generated according to the order of the overall reward value from large to small, and a set number of to-be-recommended advertisements ranked first are selected from the to-be-recommended list.

In one embodiment of the present application, the number of targeted advertisements may be multiple.

In an embodiment of the present application, a set number of advertisements to be recommended may be sent to a client where a user group administrator is located, and the user group administrator selects a target advertisement from the set number of advertisements to be recommended and forwards the target advertisement to a user group.

In one embodiment of the present application, the targeted advertisement may be pushed to the user group by the platform system where the user group is located.

In the embodiment of fig. 2, based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group, the matching probability of each advertisement to be recommended for each user is determined, the conversion success reward value corresponding to each advertisement to be recommended is obtained, the overall reward value of each advertisement to be recommended for the user group is calculated based on the matching probability and the conversion success reward value, based on the overall reward value, the target advertisement is selected from the advertisements to be recommended and recommended to the user group, the reward value gain of the user group can be considered while the proper advertisement is recommended to the user group, and further, more precise advertisement can be selected to recommend to the user group, thereby not only improving the accuracy of advertisement recommendation, but also improving the conversion rate of the advertisement, in addition, because the user obtains the reward value gain in the advertisement of the user group, the method can improve the viscosity of the users in the user group and attract the users to be continuously converted by the advertisements in the user group.

In the embodiment, the resource invested by the advertiser is the conversion success reward value, the advertiser only invests the resource under the condition that the advertiser sells the commodity, and the quantity of the sold commodity is in direct proportion to the invested quantity, so that the condition that the resource invested by the advertiser is not matched with the obtained income can be effectively avoided.

In an embodiment of the present application, in step S210 in fig. 2, determining a matching probability of each advertisement to be recommended for each user based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group may include: inputting the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group into the neural network model, and acquiring the matching probability of each advertisement to be recommended, which is output by the neural network model, aiming at each user.

In an embodiment of the application, advertisement characteristics of each advertisement to be recommended may be input to a first characteristic processing branch of the neural network model, and user characteristics of each user may be input to a second characteristic processing branch of the neural network model, where the first characteristic processing branch and the second characteristic processing branch are connected to the comprehensive processing unit, and a matching probability of each advertisement to be recommended output by the comprehensive processing unit for each user is obtained.

In one embodiment of the present application, the first feature processing branch may include: the first feature crossing module and the first feature splicing module are connected with the output end of the first feature crossing module. The first characteristic crossing module is used for carrying out characteristic crossing processing on discrete advertisement characteristics in the advertisement characteristics, and the advertisement splicing module is used for splicing the continuous advertisement characteristics in the advertisement characteristics and the advertisement crossing characteristics output by the first characteristic crossing module.

In one embodiment of the present application, the second feature processing branch may include: the time sequence feature extraction module comprises a second feature crossing module, a time sequence feature extraction module and a second feature splicing module, wherein the input end of the second feature splicing module is connected with the output end of the second feature crossing module and the output end of the time sequence feature extraction module. The second feature crossing module is used for carrying out feature crossing processing on discrete user features in the user features, the time sequence feature extraction module is used for carrying out self-attention calculation on sequence type user features in the user features, and the user splicing module is used for splicing continuous user features in the user features, the user crossing features output by the second feature crossing module and the self-attention features output by the time sequence feature extraction module.

In an embodiment of the application, the integrated processing unit may calculate a matching probability between the advertisement splicing feature output by the advertisement splicing module and the user splicing feature output by the user splicing module.

Fig. 3 schematically shows a structural diagram of a neural network model according to an embodiment of the present application, and the neural network model 300 may include a first feature processing branch 301, a second feature processing branch 302, and an integrated processing unit 303 connecting the first feature processing branch 301 and the second feature processing branch 302.

In one embodiment of the present application, the first feature processing branch 301 may include a first discrete encoding module 311, a first feature interleaving module 312 connected to an output of the first discrete encoding module 311, a first continuous encoding module 313, a first feature splicing module 314 connected to an output of the first feature interleaving module 312 and an output 313 of the first continuous encoding module, a first spatial conversion module 315 connected to an output of the first feature splicing module 314, and a first regularization module 316 connected to an output of the first spatial conversion module 315. The first discrete encoding module 311 is configured to encode discrete advertisement features in the advertisement features of each advertisement to be recommended to obtain discrete advertisement feature vectors; the first feature crossing module 312 is configured to perform feature crossing on every two discrete advertisement feature vectors in the discrete advertisement feature vectors of each advertisement to be recommended; the first continuous coding module 313 is used for coding continuous advertisement features in the advertisement features of the advertisements to be recommended to obtain continuous advertisement feature vectors; the first feature stitching module 314 is configured to stitch the continuous advertisement feature vector with the advertisement cross feature vector output by the first feature cross module 312; the first space conversion module 315 converts the advertisement splicing feature vector output by the first feature splicing module 314 into a target semantic space; the first regularization module 316 is configured to regularize the advertisement splicing feature vectors in the target semantic space.

In one embodiment of the present application, the second feature processing branch 302 may include: the time sequence feature coding module 321, a time sequence feature extracting module 322 connected to the output end of the time sequence feature coding module 321, a second discrete coding module 323, a second feature crossing module 324 connected to the output end of the second discrete coding module 323, a second continuous coding module 325, a second feature splicing module 326 connected to the output end of the time sequence feature extracting module 322, the output end of the second feature crossing module 324 and the output end of the second continuous coding module 325, a second spatial conversion module 327 connected to the output end of the second splicing feature module 326, and a second regularization module 328 connected to the output end of the second spatial conversion module 327. The time sequence feature coding module 321 is configured to perform coding processing on sequence-type user features in the user features of each user to obtain sequence-type user feature vectors; the time sequence feature extraction module 322 is configured to perform self-attention calculation on every two sequence-type user feature vectors in the sequence-type user feature vectors of each user to obtain a self-attention feature vector; the second discrete encoding module 323 is configured to perform encoding processing on discrete user features in the user features of each user to obtain discrete user feature vectors; the second feature crossing module 324 is configured to perform feature crossing on every two discrete user feature vectors in the discrete user feature vectors of each user; the second continuous encoding module 325 is configured to perform encoding processing on continuous user features in the user features of each user to obtain a continuous user feature vector; the second feature concatenation module 326 is configured to concatenate the continuous user feature vector, the self-attention feature vector, and the user cross feature vector output by the second feature cross module 324; the second space conversion module 327 converts the user splicing feature vector output by the second feature splicing module into the target semantic space; the second regularization module 328 is configured to regularize the user stitched feature vectors in the target semantic space.

In one embodiment of the present application, the first space conversion module 315 may include a plurality of fully connected layers, which can perform dimension reduction on the advertisement splicing feature vector while converting the advertisement splicing feature vector to the target semantic space. The second space conversion module 327 may include a plurality of full connection layers, which can perform dimension reduction on the user splicing feature vector while converting the user splicing feature vector to the target semantic space, so that the comprehensive processing unit calculates the matching probability between the advertisement splicing feature vector and the user splicing feature vector after dimension reduction in the same semantic space.

In an embodiment of the present application, the comprehensive processing unit 303 may be configured to calculate a feature distance between a user splicing feature vector in a target semantic space after regularization of each user and an advertisement splicing feature vector in a target semantic space after regularization of each advertisement to be recommended, and map the feature distance through an activation function, so as to obtain a matching probability of each advertisement to be recommended for each user.

In one embodiment of the present application, the structure of the neural network model can be divided into three regions, namely an input region, a projection region connected to the output end of the input region, and an output region connected to the output end of the projection region. The input region may include a first discrete encoding module 311, a first continuous encoding module 313, a timing feature encoding module 321, a second discrete encoding module 322, and a second continuous encoding module 325. The projected area may include a first feature intersection module 312, a first feature stitching module 314, a first spatial transformation module 315, a first regularization module 316, a temporal feature extraction module 322, a second feature intersection module 324, a second feature stitching module 326, a second spatial transformation module 327, and a second regularization module 328. The output section may include an integrated processing unit 303.

FIG. 4 is a schematic view ofSchematically showing a structural diagram of a first feature intersection module according to an embodiment of the present application, as shown in fig. 4, the first feature intersection module may include a plurality of Cross layers (Cross layers), and an output x of the l-th Layer_lThe corresponding formalization is expressed as follows:

wherein x is₀An input layer representing Cross Network;

a transpose representing the output of layer l-1; w is a_l-1And b_l-1Representing the weights and biases to be learned. It can be found that as the number of layers increases, the order of feature crossing increases, and the ith layer corresponds to the feature crossing of the highest order l + 1. Residual error connection is introduced, the problem of gradient disappearance is avoided to a certain extent, and the network can extract deeper features.

In this embodiment, when feature intersection is performed on discrete advertisement feature vectors, a large number of high-dimensional sparse features are faced, a large number of intersection combination modes exist, manual design feature intersection requires a large amount of manpower and trial cost, and some important intersection features are easy to miss. Based on the above problems, aiming at the aspect of explicitly constructing high-order cross features, the network structure is modified in a neural network model, various feature cross network modules are tried to be designed, and finally a first feature cross module is constructed.

In an embodiment of the present application, in the first feature crossing module, the input discrete feature vectors may be tiled and spliced, and then sent to the Cross Layer for explicit high-order crossing of features, and a high-order token may be input as a subsequent part. We used a training set of 800W samples and a test set of 200W samples in the experiment to verify the effect of the first feature crossing module and to adjust the order of feature crossing by varying the number of layers of the Cross Layer.

In one embodiment of the present application, the structure of the second feature crossing module may refer to the structure of the first feature crossing module.

Fig. 5 schematically shows a structural diagram of the time-series feature extraction module 322 according to an embodiment of the present application, and as shown in fig. 5, the time-series feature extraction module 322 may include: a Multi-headed self-attention Mechanism (MHSA) layer 3221, a first normalization (Add and Norm) layer 3222 connected to the output and input of the Multi-headed self-attention mechanism layer, a Feed Forward neural Network (FFN) connected to the output of the Multi-headed self-attention mechanism module, and a second normalization (Add and Norm) layer connected to the output and input of the Feed Forward neural Network. The multi-head self-attention mechanism layer is used for carrying out multi-head self-attention mechanism calculation on the input sequence type user characteristic vectors; the feedforward neural network layer is used for performing feedforward neural network calculation on the output of the head self-attention mechanism layer; the first normalization layer and the second normalization layer are used for performing layer normalization calculations.

In the embodiment, any two sequence type user feature vectors in the sequence type user feature vectors are directly linked through self-attention calculation, and the time sequence feature extraction module is introduced for feature extraction of sequence information, so that the relevance between the two sequence type user feature vectors can be captured, further, deeper characterization vectors can be learned, and the deep-level interdependent implicit information can be captured more easily.

In one embodiment of the present application, the neural network model may be pre-trained by: acquiring a user characteristic sample set and an advertisement characteristic sample set, and obtaining the matching probability between each user characteristic sample in the user characteristic sample set and each advertisement characteristic sample in the advertisement characteristic sample set; inputting the user characteristic samples into a first characteristic processing branch, inputting the advertisement characteristic samples into a second characteristic processing branch, and obtaining the prediction probability between each user characteristic sample in the user characteristic sample set and each advertisement characteristic sample in the advertisement characteristic sample set, which are output by the neural network model; and comparing the prediction probability with the matching probability, and if the prediction probability is inconsistent with the matching probability, adjusting the neural network model to ensure that the prediction probability is consistent with the matching probability.

In an embodiment of the application, the user characteristic sample and the advertisement characteristic sample with the matching probability reaching the set probability can be used as a positive sample, the user characteristic sample and the advertisement characteristic sample with the matching probability not reaching the set probability can be used as a negative sample, and if the ratio of the number of the positive samples to the number of the negative samples does not reach the threshold, the positive samples are copied until the ratio of the number of the positive samples to the number of the negative samples reaches the threshold.

In an embodiment of the present application, if the ratio of the number of the positive samples to the number of the negative samples does not reach the threshold, a first feature distance between each positive sample and any one of the positive samples may be calculated, and the positive samples having the feature distance smaller than the first feature distance are randomly selected as the synthesized positive samples; until the number of negative samples exceeds the sum of the number of positive samples and the number of composite positive samples by a threshold.

In one embodiment of the present application, the threshold may be set to 1: 5.

Fig. 6 schematically shows a flowchart of an advertisement recommendation method according to an embodiment of the present application, where an execution subject of the advertisement recommendation method may be a server, such as the server 103 shown in fig. 1.

Referring to fig. 6, the advertisement recommendation method at least includes steps S610 to S650:

in step S610, the basic data is accumulated.

In one embodiment of the present application, random recommendations may be made, basic data accumulation may be made to obtain positive and negative samples, and the accumulation time may be set to 3 months.

In step S620, the data is preprocessed.

In one embodiment of the present application, preprocessing may be performed on noise, missing values, logic errors, formatting errors, deduplication, etc. in the underlying data.

In one embodiment of the present application, the noise may be processed by binning, regression, or outliers, and the underlying data containing missing values may be deleted or the missing values may be interpolated.

In step S630, a sample is generated.

In one embodiment of the application, user characteristics and advertisement characteristics to be recommended may be extracted from the base data.

In one embodiment of the present application, a positive exemplar label (label) may be defined as 1 and a negative exemplar (label) may be defined as 0. The inventors have found statistically that in the positive and negative sample generation process, the positive samples are significantly less than the negative samples, and therefore are typical non-equalized data sets, so that when the positive samples are too few, we can over-sample them, and when the negative samples are too many, we can under-sample them. By comparing the experimental effects of different positive and negative sample ratios, we set the positive to negative sample ratio to 1: 5. In the process of model training and evaluation, if the number of samples is too small, the model training may be insufficient; too much sample data increases the time-consuming model training. We set the number of positive samples to 10 to 40 ten thousand by evaluation. And here to prevent the characteristic information from leaking, we select the date on which the sample was taken to be 3 days after the characteristic date.

In one embodiment of the present application, a positive sample may be analyzed and a composite sample derived from the positive sample may be added to the sample set. Calculating the distance from each positive sample to other positive samples by taking the Euclidean distance as a standard to obtain k neighbors of each positive sample; setting a sampling proportion according to the sample imbalance proportion to determine a sampling multiplying factor N, and randomly selecting a plurality of samples from k neighbors of each positive sample x, wherein the selected neighbors are assumed to be xn; for each randomly selected neighbor xn, a synthetic sample is respectively constructed with the original sample according to the following formula: xnew ═ x + rand (0, 1) × x-xn |.

With continued reference to fig. 6, in step S640, the neural network model is trained.

In an embodiment of the present application, the obtained positive and negative samples may be combined with corresponding user features and advertisement features, and input to the neural network model for training, for example: 1: user characteristic A + commodity characteristic B.

In step S650, the neural network model predicts.

In an embodiment of the present application, the neural network model in the embodiment of fig. 6 may add a calculation module at an output end of the neural network model in the embodiment of fig. 3, and the calculation module calculates an overall reward value of each advertisement to be recommended for the user group based on the matching probability and the conversion success reward value.

In an embodiment of the application, in response to an advertisement recommendation request of a user group, a user group identifier of the user group and information of each user in the user group may be obtained, user characteristics may be extracted from the user information, and the user characteristics may be input to a neural network model for prediction to obtain a target advertisement, so that the target advertisement that enables the user group to obtain a maximum overall reward value may be recommended to the user group.

In the embodiment of fig. 6, when a suitable advertisement is recommended to the user group, the reward value revenue of the user group can be considered, and the conversion of the users in the user group can be attracted, so that the revenue of the advertiser is increased, and the effectiveness of the advertiser in investing resources is improved.

In one embodiment of the present application, the training effectiveness of the neural network model may be evaluated. And (3) evaluating indexes such as the area under the curve (auc) of the new sample to the model out of the main comparison sample set off line, and evaluating indexes such as the advertisement consumption, the conversion Rate (CVR), the Click Through Rate (CTR) and the like of the new sample on the current network on line.

In one embodiment of the application, the conversion times and the overall reward value of the advertiser can be simultaneously used as evaluation indexes, and the two targets are optimized so as to further improve the effectiveness of the advertiser in investing resources.

The following describes embodiments of an apparatus of the present application, which may be used to perform the advertisement recommendation method in the above embodiments of the present application. For details that are not disclosed in the embodiments of the apparatus of the present application, please refer to the embodiments of the advertisement recommendation method described above in the present application.

FIG. 7 schematically shows a block diagram of an advertisement recommendation device according to an embodiment of the present application.

Referring to fig. 7, an advertisement recommendation apparatus 700 according to an embodiment of the present application includes a matching module 701, an obtaining module 702, a calculating module 703, and a recommending module 704.

In some embodiments of the present application, based on the foregoing solution, the matching module 701 is configured to determine, based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group, a matching probability of each advertisement to be recommended for each user; the obtaining module 702 is configured to obtain a conversion success reward value corresponding to each advertisement to be recommended; the calculation module 703 calculates the total reward value of each advertisement to be recommended for the user group based on the matching probability and the conversion success reward value; the recommendation module 704 is configured to select a targeted advertisement recommendation from the advertisements to be recommended to the user group based on the overall reward value.

In one embodiment of the present application, the calculation module 703 is configured to: calculating the product of the conversion success reward value corresponding to each advertisement to be recommended and the matching probability of each advertisement to be recommended aiming at each user to obtain the reward value of each advertisement to be recommended aiming at each user; and summing the reward values of the advertisements to be recommended aiming at the users to obtain the total reward value of the advertisements to be recommended aiming at the user group.

In one embodiment of the present application, recommendation module 704 is configured to: selecting a set number of advertisements to be recommended from the advertisements to be recommended as target advertisements according to the sequence of the overall reward value from large to small; the targeted advertisement is recommended to the user group.

In one embodiment of the present application, the matching module 701 is configured to: calculating the characteristic distance between the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended; and mapping the characteristic distance through an activation function to obtain the matching probability of each advertisement to be recommended aiming at each user.

In one embodiment of the present application, the matching module 701 is further configured to: before calculating the characteristic distance between the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended, mapping the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended to the same semantic space.

In one embodiment of the present application, the matching module 701 is configured to: acquiring the size of a user characteristic graph corresponding to the user characteristic of each user and the size of an advertisement characteristic graph corresponding to the advertisement characteristic of each advertisement to be recommended; and inputting the user characteristic of each user into a convolution layer with the convolution kernel size as the user characteristic graph size for matrix transformation, and inputting the advertisement characteristic of each advertisement to be recommended into the convolution layer with the convolution kernel size as the advertisement characteristic graph size for matrix transformation so as to convert the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended into the same semantic space.

In one embodiment of the present application, each user has a plurality of user characteristics, including: the system comprises a sequence type user characteristic used for describing the relation between user behavior and time variation, a continuous type user characteristic continuously changing along with the time variation, and a discrete type user characteristic not continuously changing along with the time variation; each advertisement to be recommended has a plurality of advertisement characteristics, and the plurality of advertisement characteristics comprise: a continuous advertisement feature that continuously changes with time, and a discrete advertisement feature that does not continuously change with time; the matching module 701 is further configured to: before mapping the user characteristics of each user and the advertisement characteristics of each advertisement to be recommended to the same semantic space, splicing the multiple user characteristics of each user, and splicing the multiple advertisement characteristics of each advertisement to be recommended.

In one embodiment of the present application, the matching module 701 is configured to: inputting the advertisement characteristics of each advertisement to be recommended to a first characteristic processing branch of the neural network model, and inputting the user characteristics of each user to a second characteristic processing branch of the neural network model, wherein the first characteristic processing branch and the second characteristic processing branch are connected to the comprehensive processing unit; and acquiring the matching probability of each advertisement to be recommended, which is output by the comprehensive processing unit, aiming at each user.

In one embodiment of the present application, the first feature processing branch comprises: the first characteristic cross module and the first characteristic splicing module are connected with the output end of the first characteristic cross module; the first characteristic crossing module is used for carrying out characteristic crossing processing on discrete advertisement characteristics in the advertisement characteristics, and the advertisement splicing module is used for splicing the continuous advertisement characteristics in the advertisement characteristics and the advertisement crossing characteristics output by the first characteristic crossing module; the second feature processing branch includes: the time sequence feature extraction module is connected with the output end of the first feature crossing module; the second feature crossing module is used for carrying out feature crossing processing on discrete user features in the user features, the time sequence feature extraction module is used for carrying out self-attention calculation on sequence type user features in the user features, and the user splicing module is used for splicing continuous user features in the user features, the user crossing features output by the second feature crossing module and the self-attention features output by the time sequence feature extraction module.

In one embodiment of the present application, the neural network model is pre-trained by: acquiring a user characteristic sample set and an advertisement characteristic sample set, and obtaining the matching probability between each user characteristic sample in the user characteristic sample set and each advertisement characteristic sample in the advertisement characteristic sample set; inputting the user characteristic samples into a first characteristic processing branch, inputting the advertisement characteristic samples into a second characteristic processing branch, and obtaining the prediction probability between each user characteristic sample in the user characteristic sample set and each advertisement characteristic sample in the advertisement characteristic sample set, which are output by the neural network model; and comparing the prediction probability with the matching probability, and if the prediction probability is inconsistent with the matching probability, adjusting the neural network model to ensure that the prediction probability is consistent with the matching probability.

In one embodiment of the present application, the advertisement recommendation device 700 further includes: a synthesis module configured to: taking the user characteristic sample and the advertisement characteristic sample with the matching probability reaching the set probability as positive samples; taking the user characteristic sample and the advertisement characteristic sample with the matching probability not reaching the set probability as negative samples; if the ratio of the number of the positive samples to the number of the negative samples does not reach a threshold value, calculating a first characteristic distance between each positive sample and any one positive sample, and randomly selecting the positive sample with the characteristic distance smaller than the first characteristic distance as a synthesized positive sample; until the number of negative samples exceeds the sum of the number of positive samples and the number of composite positive samples by a threshold.

As will be appreciated by one skilled in the art, aspects of the present application may be embodied as a system, method or program product. Accordingly, various aspects of the present application may be embodied in the form of: an entirely hardware embodiment, an entirely software embodiment (including firmware, microcode, etc.) or an embodiment combining hardware and software aspects that may all generally be referred to herein as a "circuit," module "or" system.

An electronic device 80 according to this embodiment of the present application is described below with reference to fig. 8. The electronic device 80 shown in fig. 8 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present application.

As shown in fig. 8, the electronic device 80 is in the form of a general purpose computing device. The components of the electronic device 80 may include, but are not limited to: the at least one processing unit 81, the at least one memory unit 82, a bus 83 connecting different system components (including the memory unit 82 and the processing unit 81), and a display unit 84.

Wherein the storage unit stores program code that can be executed by the processing unit 81 such that the processing unit 81 performs the steps according to various exemplary embodiments of the present application described in the section "example methods" above in this specification.

The storage unit 82 may include readable media in the form of volatile storage units, such as a random access storage unit (RAM)821 and/or a cache storage unit 822, and may further include a read only storage unit (ROM) 823.

The storage unit 82 may also include a program/utility 824 having a set (at least one) of program modules 825, such program modules 825 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.

Bus 83 may be any of several types of bus structures including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

The electronic device 80 may also communicate with one or more external devices (e.g., keyboard, pointing device, bluetooth device, etc.), with one or more devices that enable a user to interact with the electronic device 80, and/or with any devices (e.g., router, modem, etc.) that enable the electronic device 80 to communicate with one or more other computing devices. Such communication may be through input/output (I/O) interfaces 85. Also, the electronic device 80 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet) via the network adapter 86. As shown, the network adapter 86 communicates with the other modules of the electronic device 80 via the bus 83. It should be understood that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the electronic device 80, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.

Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein may be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solution according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a usb disk, a removable hard disk, etc.) or on a network, and includes several instructions to make a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) execute the method according to the embodiments of the present application.

There is also provided, in accordance with an embodiment of the present application, a computer-readable storage medium having stored thereon a program product capable of implementing the above-described method of the present specification. In some possible embodiments, various aspects of the present application may also be implemented in the form of a program product comprising program code for causing a terminal device to perform the steps according to various exemplary embodiments of the present application described in the "exemplary methods" section above of this specification, when the program product is run on the terminal device.

According to one embodiment of the present application, a program product for implementing the above method may employ a portable compact disc read only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

A computer readable signal medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations of the present application may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., through the internet using an internet service provider).

Furthermore, the above-described figures are merely schematic illustrations of processes involved in methods according to exemplary embodiments of the present application, and are not intended to be limiting. It will be readily understood that the processes shown in the above figures are not intended to indicate or limit the chronological order of the processes. In addition, it is also readily understood that these processes may be performed synchronously or asynchronously, e.g., in multiple modules.

It will be understood that the present application is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the application is limited only by the appended claims.

Claims

1. An advertisement recommendation method, comprising:

determining the matching probability of each advertisement to be recommended aiming at each user based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group;

acquiring a conversion success reward value corresponding to each advertisement to be recommended;

calculating the total reward value of each advertisement to be recommended aiming at the user group based on the matching probability and the conversion success reward value;

and selecting target advertisements from the advertisements to be recommended to recommend to the user group based on the overall reward value.

2. The advertisement recommendation method according to claim 1, wherein the calculating an overall reward value of each advertisement to be recommended for the user group based on the matching probability and the conversion success reward value comprises:

calculating the product of the conversion success reward value corresponding to each advertisement to be recommended and the matching probability of each advertisement to be recommended aiming at each user to obtain the reward value of each advertisement to be recommended aiming at each user;

and summing the reward values of the advertisements to be recommended aiming at the users to obtain the total reward value of the advertisements to be recommended aiming at the user group.

3. The advertisement recommendation method of claim 1, wherein the selecting a target advertisement recommendation from the advertisements to be recommended to the user group based on the overall reward value comprises:

selecting a set number of advertisements to be recommended from the advertisements to be recommended as the target advertisements according to the sequence of the overall reward value from large to small;

recommending the targeted advertisement to the user group.

4. The advertisement recommendation method according to claim 1, wherein the determining the matching probability of each advertisement to be recommended for each user based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group comprises:

calculating the characteristic distance between the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended;

and mapping the characteristic distance through an activation function to obtain the matching probability of each advertisement to be recommended aiming at each user.

5. The advertisement recommendation method according to claim 4, wherein before calculating the feature distance between the user feature of each user and the advertisement feature of each advertisement to be recommended, the method further comprises:

and mapping the user characteristics of each user and the advertisement characteristics of each advertisement to be recommended to the same semantic space.

6. The advertisement recommendation method according to claim 5, wherein said mapping the user characteristics of each user and the advertisement characteristics of each advertisement to be recommended to the same semantic space comprises:

acquiring the size of a user characteristic graph corresponding to the user characteristic of each user and the size of an advertisement characteristic graph corresponding to the advertisement characteristic of each advertisement to be recommended;

and performing matrix transformation on the convolution layer with the user characteristic input convolution kernel size of each user as the user characteristic graph size, and performing matrix transformation on the convolution layer with the advertisement characteristic input convolution kernel size of each advertisement to be recommended as the advertisement characteristic graph size so as to convert the user characteristic of each user and the advertisement characteristic of each advertisement to be recommended into the same semantic space.

7. The advertisement recommendation method of claim 5, wherein each user has a plurality of user characteristics, the plurality of user characteristics comprising: the system comprises a sequence type user characteristic used for describing the relation between user behavior and time variation, a continuous type user characteristic continuously changing along with the time variation, and a discrete type user characteristic not continuously changing along with the time variation;

each advertisement to be recommended has a plurality of advertisement characteristics, wherein the plurality of advertisement characteristics comprise: a continuous advertisement feature that continuously changes with time, and a discrete advertisement feature that does not continuously change with time;

before mapping the user characteristics of each user and the advertisement characteristics of each advertisement to be recommended to the same semantic space, the advertisement recommendation method further includes: and splicing the multiple user characteristics of each user, and splicing the multiple advertisement characteristics of each advertisement to be recommended.

8. The advertisement recommendation method according to claim 7, wherein:

and splicing the multiple user characteristics of each user, including: performing self-attention calculation on every two sequence type user characteristics of each user to obtain self-attention characteristics; performing feature cross processing on every two discrete user features of each user to obtain user cross features; splicing the continuous user features, the self-attention features and the user cross features of the users;

splicing the various advertisement characteristics of the advertisements to be recommended, wherein the splicing process comprises the following steps: performing feature cross processing on every two discrete advertisement features of each advertisement to be recommended to obtain advertisement cross features; and splicing the continuous advertisement characteristics and the advertisement cross characteristics of the advertisements to be recommended.

9. The advertisement recommendation method according to claim 1, wherein the determining the matching probability of each advertisement to be recommended for each user based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group comprises:

inputting the advertisement characteristics of each advertisement to be recommended to a first characteristic processing branch of a neural network model, and inputting the user characteristics of each user to a second characteristic processing branch of the neural network model, wherein the first characteristic processing branch and the second characteristic processing branch are connected to a comprehensive processing unit;

and acquiring the matching probability of each advertisement to be recommended, which is output by the comprehensive processing unit, aiming at each user.

10. The advertisement recommendation method according to claim 9, wherein the first feature processing branch comprises: the first characteristic cross module and the first characteristic splicing module are connected with the output end of the first characteristic cross module;

the first feature crossing module is used for performing feature crossing processing on discrete advertisement features in the advertisement features, and the advertisement splicing module is used for splicing continuous advertisement features in the advertisement features and the advertisement crossing features output by the first feature crossing module;

the second feature processing branch comprises: the time sequence feature extraction module is connected with the output end of the second feature crossing module;

the second feature crossing module is configured to perform feature crossing processing on discrete user features in the user features, the time sequence feature extraction module is configured to perform self-attention calculation on sequence-type user features in the user features, and the user splicing module is configured to splice continuous user features in the user features, the user crossing features output by the second feature crossing module, and the self-attention features output by the time sequence feature extraction module.

11. The advertisement recommendation method of claim 10, wherein the neural network model is pre-trained by:

acquiring a user characteristic sample set and an advertisement characteristic sample set, and matching probabilities between each user characteristic sample in the user characteristic sample set and each advertisement characteristic sample in the advertisement characteristic sample set;

inputting the user characteristic samples into the first characteristic processing branch, inputting the advertisement characteristic samples into the second characteristic processing branch, and obtaining the prediction probability between each user characteristic sample in the user characteristic sample set and each advertisement characteristic sample in the advertisement characteristic sample set output by the neural network model;

comparing the predicted probability with the matching probability, and if not, adjusting the neural network model so that the predicted probability is consistent with the matching probability.

12. The advertisement recommendation method according to claim 11, wherein before inputting the user feature sample to the first feature processing branch and inputting the advertisement feature sample to the second feature processing branch, the method comprises:

taking the user characteristic sample and the advertisement characteristic sample with the matching probability reaching the set probability as positive samples;

taking the user characteristic sample and the advertisement characteristic sample with the matching probability not reaching the set probability as negative samples;

if the ratio of the number of the positive samples to the number of the negative samples does not reach a threshold value, calculating a first characteristic distance between each positive sample and any one of the positive samples, and randomly selecting the positive sample with the characteristic distance smaller than the first characteristic distance as a synthesized positive sample;

until the number of negative samples reaches the threshold in a ratio of the number of positive samples to the number of composite positive samples.

13. An advertisement recommendation apparatus, comprising:

the matching module is configured to determine the matching probability of each advertisement to be recommended aiming at each user based on the advertisement characteristics of each advertisement to be recommended and the user characteristics of each user in the user group;

the acquisition module is configured to acquire the conversion success reward value corresponding to each advertisement to be recommended;

the calculation module is used for calculating the total reward value of each advertisement to be recommended aiming at the user group based on the matching probability and the conversion success reward value;

and the recommending module is configured to select a target advertisement from the advertisements to be recommended and recommend the target advertisement to the user group based on the overall reward value.

14. An electronic device, comprising:

a memory storing computer readable instructions;

a processor to read computer readable instructions stored by the memory to perform the method of any of claims 1-12.

15. A computer program medium having computer readable instructions stored thereon which, when executed by a processor of a computer, cause the computer to perform the method of any one of claims 1-12.