CN116385076A

CN116385076A - Advertisement recommendation method, advertisement recommendation device, terminal equipment and computer readable storage medium

Info

Publication number: CN116385076A
Application number: CN202310319129.8A
Authority: CN
Inventors: 于长颖; 朱火庚
Original assignee: Guangzhou Tidong Technology Co ltd
Current assignee: Guangzhou Tidong Technology Co ltd
Priority date: 2023-03-28
Filing date: 2023-03-28
Publication date: 2023-07-04

Abstract

The application provides an advertisement recommendation method, an advertisement recommendation device, a terminal device and a computer readable storage medium, wherein advertisement click and exposure behavior data of a user needing advertisement recommendation are obtained; recall advertisements according to a preset advertisement recall logic rule to form an initial advertisement set; inputting advertisement clicking and exposing behavior data and an initial advertisement set into a pre-trained LightGBM model, and calculating clicking rate scores of advertisements in the initial advertisement set; a preset number of advertisement recommendations from the initial advertisement set are selected to the user based on each click-through rate score. According to the method, the click rate score of the advertisement can be accurately calculated by adopting the LightGBM model based on the advertisement click and exposure behavior data of the user, so that the accuracy of recommending the advertisement can be provided, and the time for screening the advertisement by the user is greatly reduced. In addition, by adopting the method, different advertisement recommendation can be carried out on different users, and the requirements of different clients are met.

Description

Advertisement recommendation method, advertisement recommendation device, terminal equipment and computer readable storage medium

Technical Field

The present disclosure relates to the field of advertisement recommendation technologies, and in particular, to an advertisement recommendation method, an advertisement recommendation device, a terminal device, and a computer readable storage medium.

Background

An advertisement creative customizing platform is a matching platform for providing advertisement commodity creative customizing service for advertisers. When the user needs to purchase the advertisement, the user can enter the advertisement creative customization platform, the commodity platform can display different commodities for the client to select, the user selects the advertisement to click into a detail page, and finally the conversion of ordering is completed. However, there are often a large number of advertisements in the advertisement creative customization platform, and users need to browse the large number of advertisements to select the content needed by the users, so that the operation is very troublesome. Although the current advertising creative customization platform has advertising recommendation capability, the current recommended advertisements are often not needed by users, namely, the recommendation accuracy is low and the effect is poor.

Disclosure of Invention

In view of this, an advertisement recommendation method, apparatus, terminal device and computer readable storage medium are provided in the embodiments of the present application to overcome the problem of low advertisement recommendation accuracy and poor effect in the prior art.

In a first aspect, an embodiment of the present application provides an advertisement recommendation method, including:

acquiring advertisement clicking and exposure behavior data of a user needing advertisement recommendation; the advertisement clicking and exposing behavior data are data formed by clicking advertisements when a user enters an advertisement creative customizing platform;

recall advertisements according to a preset advertisement recall logic rule to form an initial advertisement set;

inputting the advertisement clicking and exposing behavior data and the initial advertisement set into a pre-trained LightGBM model, and calculating clicking rate scores of advertisements in the initial advertisement set;

and selecting a preset number of advertisement recommendations from the initial advertisement set to a user based on each click rate score.

In a second aspect, an embodiment of the present application provides an advertisement recommendation apparatus, including:

the data acquisition module is used for acquiring advertisement clicking and exposure behavior data of a user needing advertisement recommendation; the advertisement clicking and exposing behavior data are data formed by clicking advertisements when a user enters an advertisement creative customizing platform;

the advertisement recall module is used for recalling advertisements according to a preset advertisement recall logic rule so as to form an initial advertisement set;

the click rate scoring module is used for inputting the advertisement click and exposure behavior data and the initial advertisement set into a pre-trained LightGBM model, and calculating click rate scores of advertisements in the initial advertisement set;

and the advertisement recommendation module is used for selecting a preset number of advertisement recommendations from the initial advertisement set to users based on the click rate scores.

In a third aspect, an embodiment of the present application provides a terminal device, including: a memory; one or more processors coupled with the memory; one or more applications, wherein the one or more applications are stored in memory and configured to be executed by the one or more processors, the one or more applications configured to perform the advertisement recommendation method provided in the first aspect described above.

In a fourth aspect, embodiments of the present application provide a computer readable storage medium having program code stored therein, the program code being executable by a processor to perform the advertisement recommendation method provided in the first aspect.

The advertisement recommendation method, the advertisement recommendation device, the terminal equipment and the computer readable storage medium provided by the embodiment of the application firstly acquire advertisement click and exposure behavior data of a user needing advertisement recommendation; the advertisement clicking and exposing behavior data is data formed by clicking advertisements when a user enters an advertisement creative customizing platform; then recalling advertisements according to a preset advertisement recall logic rule to form an initial advertisement set; inputting advertisement clicking and exposing behavior data and an initial advertisement set into a pre-trained LightGBM model, and calculating clicking rate scores of advertisements in the initial advertisement set; and finally, selecting a preset number of advertisement recommendations from the initial advertisement set to the user based on each click rate score.

According to the advertisement recommendation method provided by the embodiment of the application, a LightGBM model (namely a click rate prediction model) is adopted, and the click rate of advertisements in an initial advertisement set which is preliminarily recalled is scored according to advertisement click and exposure behavior data of a user, wherein the click rate score represents the probability that the user clicks a certain advertisement to place an order, and when the value is higher, the probability that the user clicks to place an order is higher, otherwise, the probability that the user clicks to place an order is lower; advertisements are then recommended to the user based on the click through rate scores. According to the method, the click rate score of the advertisement can be accurately calculated by adopting the LightGBM model based on the advertisement click and exposure behavior data of the user, so that the accuracy of recommending the advertisement can be provided, and the time for screening the advertisement by the user is greatly reduced. In addition, by adopting the method, different advertisement recommendation can be carried out on different users, and the requirements of different clients are met.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present application, and that other drawings may be obtained according to the provided drawings without inventive effort to a person skilled in the art.

Fig. 1 is an application scenario schematic diagram of an advertisement recommendation method provided in an embodiment of the present application;

FIG. 2 is a flowchart illustrating an advertisement recommendation method according to an embodiment of the present application;

FIG. 3 is a schematic diagram of an advertisement recommendation device according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of a terminal device provided in an embodiment of the present application;

fig. 5 is a schematic structural diagram of a computer-readable storage medium provided in an embodiment of the present application.

Detailed Description

The following description of the technical solutions in the embodiments of the present application will be clear and complete, and it is obvious that the described embodiments are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.

For a more detailed description of the present application, an advertisement recommendation method, apparatus, terminal device, and computer-readable storage medium provided in the present application are specifically described below with reference to the accompanying drawings.

Referring to fig. 1, fig. 1 shows a schematic diagram of an application scenario of an advertisement recommendation method provided in an embodiment of the present application, where the application scenario includes a terminal device 100 provided in an embodiment of the present application, and the terminal device 100 may be various electronic devices (such as a structure diagram of 102, 104, 106, and 108) with a display screen, including, but not limited to, a smart phone and a computer device, where the computer device may be at least one of a desktop computer, a portable computer, a laptop computer, a tablet computer, and the like. The terminal device 100 may refer broadly to one of a plurality of terminal devices, and the present embodiment is illustrated with the terminal device 100 only. Those skilled in the art will appreciate that the number of terminal devices described above may be greater or lesser. For example, the number of the terminal devices may be only several, or the number of the terminal devices may be tens or hundreds, or more, and the number and types of the terminal devices are not limited in the embodiment of the present application. The terminal device 100 may be configured to perform an advertisement recommendation method provided in an embodiment of the present application.

In an optional implementation manner, the application scenario may further include a server in addition to the terminal device 100 provided in the embodiment of the present application, where a network is disposed between the server and the terminal device. The network is used as a medium for providing a communication link between the terminal device and the server. The network may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

It should be understood that the number of terminal devices, networks and servers is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation. For example, the server may be a server cluster formed by a plurality of servers. Wherein the terminal device interacts with the server through the network to receive or send messages and the like. The server may be a server providing various services. Wherein the server may be adapted to perform the steps of an advertisement recommendation method provided in embodiments of the present application. In addition, when the terminal device executes an advertisement recommendation method provided in the embodiment of the present application, a part of steps may be executed in the terminal device, and a part of steps may be executed in the server, which is not limited herein.

Based on the above, an advertisement recommendation method is provided in the embodiments of the present application. Referring to fig. 2, fig. 2 shows a flowchart of an advertisement recommendation method provided in an embodiment of the present application, and the method is applied to the terminal device in fig. 1 for illustration, and includes the following steps:

step S110, advertisement click and exposure behavior data of a user needing advertisement recommendation are obtained.

The advertisement clicking and exposing behavior data is data formed by clicking advertisements when a user enters an advertisement creative customizing platform.

In one embodiment, in performing step S110, acquiring advertisement click and exposure behavior data of a user who needs advertisement recommendation includes: and acquiring advertisement click and exposure behavior data of a user needing advertisement recommendation from a log system by adopting a buried point mode.

Specifically, the advertisement recommendation process is to collect advertisement click and exposure behavior data when a user enters an advertisement creative customization platform, then to predict click rate scores of advertisements in a recalled initial advertisement set according to the data, and finally to select advertisements meeting certain conditions according to the click rate score ranking and push the advertisements to the user. In this embodiment, a search execution pipeline SearchPipeline may be defined on the ad creative platform to control the overall flow, where the pipeline works by running the following 4 search executors in parallel, ensuring a high throughput of data and a high availability stability of the system. Specifically, a responsibility chain mode is adopted, original search parameters and results of each stage are stored through self-defining search execution pipeline context, and then a processing logic SearchHandler of a single search stage is sequentially executed by the search execution pipeline SearchPipeline. In addition, the advertisement recommendation architecture can be configured on an advertisement creative platform, and comprises 4 search executors, namely a commodity recall executor SpuRecall Handler, a commodity algorithm fine-ranking executor SpuSortHandler, a commodity packaging processor SearchPackHandler (SpuPacker) and a record search keyword SearchRecordHandler, wherein the processors are all software modules which are written at one time and run everywhere for reducing cost and facilitating iteration, and can attract excellent engineers all over the world through an open source to optimize together, so that the performance is improved.

Advertisement click and exposure behavior data refers to behavior data generated when a user (also called an advertiser) enters an advertisement creative customization platform and the platform browses search clicks to download advertisements, and comprises advertisement exposure behavior data and advertisement click data, wherein each piece of data represents one advertisement exposure or click behavior.

The advertisement clicking and exposing behavior data can be collected through a front-end embedded point and obtained from a log system.

Step S120, recalling advertisements according to a preset advertisement recall logic rule to form an initial advertisement set.

In one embodiment, in executing step S120, recalling advertisements according to a preset advertisement recall logic rule, comprising: the advertisement is recalled according to the elastic search index recall rule, and according to the integrated scoring rule of advertisement name, advertisement creator name, and advertisement query sales.

The preset advertisement recall logic rule refers to a preset advertisement recall rule, and is used for carrying out preliminary advertisement recall on the advertisement creative customization platform. Alternatively, the preset advertisement recall logic may be an elastic search index recall rule, and a comprehensive scoring rule based on advertisement name, advertisement creator name, and advertisement query sales.

Further, the elastic search supports full text searching of documents, and an inverted index is established for each field of data, and is an index data structure used by the elastic search for full text searching.

Step S130, inputting advertisement clicking and exposing behavior data and an initial advertisement set into a pre-trained LightGBM model, and calculating clicking rate scores of advertisements in the initial advertisement set.

Specifically, the LightGBM model is an industrial framework for realizing the GBDT algorithm, has the advantages of faster training speed, lower memory consumption, better accuracy, distributed support and the like, and has good application in click rate prediction. In the field of machine learning, GBDT is a very good algorithm model, and is trained iteratively by using the thought of a weak classifier (decision tree), so that an optimal model is obtained, and the model has the advantages of good performance, difficulty in overfitting and the like. GBDT application in the internet world is very popular, and is used for tasks such as multi-classification, click rate prediction, search ranking and the like, and most of excellent schemes of competition on kagle are based on GBDT.

LightGBM mainly fuses 2 algorithms: a single-Sampling algorithm (GOSS) and a feature bundling algorithm (Exclusive Feature Bunding, EFB). Single-sided Gradient Sampling Gradient-based One-Side Sampling (GOSS): the use of GOSS can reduce a large number of data instances with only small gradients, so that only the remaining data with high gradients can be used when calculating the information gain, and compared with XGBoost, the method has the advantage of saving a lot of time and space overhead. Mutually exclusive feature bundle Exclusive Feature Bundling (EFB): the EFB can bind a plurality of mutually exclusive features into one feature, thus achieving the purpose of dimension reduction.

In this embodiment, the LightGBM model is employed and click rate scoring is performed on advertisements in the initial advertisement set based on the advertisement click and exposure behavior data.

Step S140, selecting a preset number of advertisement recommendations from the initial advertisement set to the user based on each click-through rate score.

In one embodiment, selecting a preset number of advertisement recommendations from an initial advertisement set to a user based on each click-through rate score comprises: selecting advertisements with click rate scores greater than a preset threshold from the initial advertisement set to form a candidate advertisement set; a preset number of advertisement recommendations is selected from the candidate advertisement set to the user.

The preset number is preset data, and can be any positive integer, such as 1, 5, 10, 20, and the like, and specifically can be determined according to the size of the display page of the advertisement creative customization platform.

The preset threshold is a preset value, and may be any positive number. In practice, the method can be determined according to specific requirements, and in general, the larger the preset threshold value is, the higher the advertisement accuracy recommended to the user is, and the lower the advertisement accuracy recommended to the user by the anti-regularization is.

According to the advertisement recommendation method provided by the embodiment of the application, advertisement clicking and exposure behavior data of a user needing advertisement recommendation are firstly obtained; the advertisement clicking and exposing behavior data is data formed by clicking advertisements when a user enters an advertisement creative customizing platform; then recalling advertisements according to a preset advertisement recall logic rule to form an initial advertisement set; inputting advertisement clicking and exposing behavior data and an initial advertisement set into a pre-trained LightGBM model, and calculating clicking rate scores of advertisements in the initial advertisement set; and finally, selecting a preset number of advertisement recommendations from the initial advertisement set to the user based on each click rate score.

Further, an embodiment for training the LightGBM model is presented, as detailed below:

in one embodiment, training the LightGBM model includes:

step S1, acquiring advertisement click and exposure behavior data samples, and dividing the advertisement click and exposure behavior data samples into a training set and a testing set.

Specifically, the advertisement click and exposure behavior data samples may be historical advertisement click and exposure behavior data. Typically, historical ad exposure and click behavior data is developed by different users (i.e., advertisers) into the ad creative customization platform over a period of time (e.g., 2021, 8, and 2021, 11, end).

Alternatively, the advertisement click and exposure behavior data samples may be Hive format data. Further, the data sample of the advertisement clicking and exposing behavior in the Hive format can be processed by using a Dorado platform through HiveSQL sentences, and the offline data is processed by using the Dorado under the conditions that the data sample size is large and the timeliness requirement of the data processing is not high. The Dorado is a big data research and development platform integrating the functions of data integration, data development, task scheduling, operation and maintenance management, data analysis and the like, provides a one-stop big data research and development solution, helps business departments to construct a plurality of bins, comprises ETL development, data analysis and exploration, simply and efficiently constructs own data center, and is focused on mining and exploration of data values.

While the specific process of using dordao to process offline data is to convert a batch of pending Hive tables (in which live format ad click and exposure behavior data samples are stored) into HSQL task processing data, and then process the Hive data into a new Hiva table.

The samples may be finally stored in a table named mark_spu_card_ctr_samples after construction, and each sample data of the table records exposure click data of a user (i.e. advertiser) on an advertisement under the flag of an advertisement creator, and the sample data is positive if the user (i.e. advertiser) exposes and clicks on the advertisement, and is negative if the commodity is exposed and not clicked.

In this embodiment, historical advertisement exposure and click behavior data from month 2021, 8, to month 2021, 11, bottom can be selected, yielding a total of 1 273 186 samples. These samples may in turn form training and testing sets, as shown in table 1 for details:

table 1 is a sample of advertisement click and exposure behavior data

Data set	Time	Number of samples/bar	Number of positive samples/bar
				Training set	20210812-20211107	1026904	40396
Test set	20211108-20211130	246282	6845

And S2, inputting advertisement click and exposure behavior data samples in the training set into the LightGBM model, and adjusting parameters of the LightGBM model until convergence to obtain an initial LightGBM model.

In one embodiment, step S2 is performed to adjust parameters of the LightGBM model, including: the LGBMClassifier function is used to adjust the parameters of the LightGBM model.

And S3, testing the initial LightGBM model by adopting advertisement click and exposure behavior data samples in the test set, evaluating a feature_importance feature tree of the initial LightGBM model by adopting feature data, and obtaining a trained LightGBM model when the test and the evaluation pass.

Specifically, lightGBM mainly fuses 2 algorithms: a single-Sampling algorithm (GOSS) and a feature bundling algorithm (Exclusive Feature Bunding, EFB). The GOSS orders the absolute gradient values of the samples at the time of sampling, retains a×100% of large gradient samples before selection, randomly samples the remaining (1-a) ×100% of small gradient samples, randomly selects b× (1-a) ×100% of data, and multiplies (1-n)/m as information gain. A new weak learner is learned using this method, and repeated until convergence. The benefit of such an optimization is that large gradient samples have more information gain, and that the whole training can be positively acted upon without giving excessive attention to small samples. The GOSS algorithm reduces the time complexity by reducing the number of samples, while the EFB algorithm reduces the complexity by considering the reduction of the number of features. Generally, the data used will not take a value of 0 at the same time, i.e. there is a mutual exclusion feature. The EFB algorithm realizes dimension reduction by binding the mutual exclusion features to reduce the number of the mutual exclusion features. Selecting mutually exclusive features as binding objects can avoid losing information, and if 2 features are not completely mutually exclusive, the degree of non-mutual exclusion between features can be measured by introducing a conflict ratio. If the conflict ratio value is small, even if the 2 features which are not completely mutually exclusive are bundled, the final precision is not affected. Training the LightGBM model based on this mainly comprises two steps: (1) importing the training set and the test set. (2) optimizing the model parameters. The training set is mainly used for training the model to obtain an initial LightGBM model, and the testing set is mainly used for testing the initial LightGBM model, and when the testing result is optimal or meets the preset condition, the final LightGBM model is obtained.

Further, the test set employed primarily to test the initial LightGBM model may be a confusion matrix test (or evaluation). The confusion matrix is a situation analysis table for summarizing the prediction results of the classification model in data science, data analysis and machine learning, and records in the data set are summarized in a matrix form according to 2 standards of classification judgment made by the real category and the classification model. The confusion matrix for test samples after modeling using LightGBM is shown in table 2.

Table 2 shows confusion matrix for test set LightGBM model

As can be seen from table 2, of 246 282 test samples, 6 845 samples were actually clicked, 4702 samples were actually positive (TP), and 2 samples were actually negative (FN) data; the true click-free data were 239 437, 58 False Positives (FP) and 181206 True Negatives (TN). The True Positive Rate (TPR) can be calculated therefrom to evaluate the sensitivity of the model, and the True Negative Rate (TNR) can be used to evaluate the specificity of the model. Where tpr=tp/(tp+fn) =4702/(4702+2143) = 0.68692; tnr=tn/(tn+fp) = 181206/(181206+58231) = 0.75680.

Alternatively, the parameters of the LightGBM model may include learner type (Objective), learning rate (learning_rate), feature ratio (feature_fraction), sample ratio (bagging_fraction), maximum depth of tree (max_depth), most leaf tree on one tree (num_leave), and metric (metric).

Wherein, the value of the learner type (object) can be a binary; the Learning rate (learning_rate) may have a value of 0.1; the Feature ratio (feature_fraction) may take a value of 0.7; the value of the sample ratio (bagging_fraction) may be 1; the maximum depth (max_depth) of the tree may take a value of 8; the value of the most leaf tree (num_leave) on one tree may be 40; and the metric (metric) may be cross_entcopy, auc.

In addition, the LGBMClassifier function in the LightGBM package in python can be used to adjust the parameters of the LightGBM model.

In order to accurately predict click rate scores for advertisements using the LightGBM model, feature data is required to evaluate them. In the internet era of information transparency, the user characteristic behavior is particularly critical to the update iteration of internet products, so in this embodiment, 3 characteristic data may be adopted, which are user side characteristic data, advertisement side characteristic data, and user and advertisement cross characteristic data, respectively. The feature data can be Hive format data, the processing method can refer to the advertisement click and exposure behavior data sample processing mode, namely, the feature big data can be processed offline through an HSQL task, and the 3 pieces of Hive table feature data are processed to the final click rate Hive table ctr_mark_label_and_features, and the total is 166 pieces of feature data. After obtaining the feature data, 166 feature data may be input into a feature_importance feature tree at the LightGBM training site, and specific feature examples (ordered according to feature importance) of 3 features are obtained through the importance ranking of the feature tree, as shown in table 3.

Table 3 shows the importance ranking of the feature data

From Table 3, it can be seen that the most important feature in the advertisement dimension is the CTR of the advertisement at the platform. Because CTR is used as a user behavior index, the dominant factor is the user and the next is the advertisement dimension information. In summary, it is very accurate to calculate the click rate scores of the advertisements in the initial advertisement set from the advertisement click and exposure behavior data using the LightGBM model.

It should be understood that, although the steps in the flowcharts of fig. 2 and 5 are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps of fig. 2 and 5 may include multiple sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor do the order in which the sub-steps or stages are performed necessarily occur in sequence, but may be performed alternately or alternately with at least a portion of the sub-steps or stages of other steps or steps.

The embodiment disclosed in the application describes an advertisement recommendation method in detail, and the method disclosed in the application can be implemented by using various types of equipment, so that the application also discloses an advertisement recommendation device corresponding to the method, and specific embodiments are given below for detailed description.

Referring to fig. 3, an advertisement recommendation device disclosed in an embodiment of the present application mainly includes:

a data acquisition module 310, configured to acquire advertisement click and exposure behavior data of a user who needs advertisement recommendation; the advertisement clicking and exposing behavior data is data formed by clicking advertisements when a user enters an advertisement creative customizing platform;

an advertisement recall module 320, configured to recall advertisements according to a preset advertisement recall logic rule to form an initial advertisement set;

the click rate scoring module 330 is configured to input the advertisement click and exposure behavior data and the initial advertisement set into a pre-trained LightGBM model, and calculate a click rate score of each advertisement in the initial advertisement set;

the advertisement recommendation module 340 is configured to select a preset number of advertisement recommendations from the initial advertisement set to the user based on each click-through rate score.

In one embodiment, the advertisement recommendation module 340 is configured to select advertisements from the initial advertisement set that have click-through rates greater than a preset threshold to form a candidate advertisement set; a preset number of advertisement recommendations is selected from the candidate advertisement set to the user.

In one embodiment, an apparatus comprises:

the sample acquisition module is used for acquiring advertisement click and exposure behavior data samples and dividing the advertisement click and exposure behavior data samples into a training set and a testing set;

the model training module is used for inputting advertisement click and exposure behavior data samples in the training set into the LightGBM model, and adjusting parameters of the LightGBM model until convergence so as to obtain an initial LightGBM model;

the model test evaluation module is used for testing the initial LightGBM model by adopting advertisement click and exposure behavior data samples in the test set, evaluating the feature-input feature tree of the initial LightGBM model by adopting feature data, and obtaining the trained LightGBM model when the test and the evaluation pass.

In one embodiment, the model training module is configured to adjust parameters of the LightGBM model using an LGBMClassifier function.

In one embodiment, advertisement recall module 320 is configured to recall advertisements according to the elastic search index recall rules, as well as according to the composite scoring rules of advertisement name, advertisement creator name, and advertisement query sales.

In one embodiment, the data acquisition module 310 is configured to acquire advertisement click and exposure behavior data of a user who needs advertisement recommendation in a buried point manner and from a log system.

In one embodiment, the feature data includes user-side feature data, advertisement-side feature data, and user-and advertisement-crossing feature data.

For specific limitations of the advertisement recommendation apparatus, reference may be made to the limitations of the method described above, and no further description is given here. Each of the modules in the above-described apparatus may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or independent of a processor in the terminal device, or may be stored in software in a memory in the terminal device, so that the processor may call and execute operations corresponding to the above modules.

Referring to fig. 4, fig. 4 shows a block diagram of a terminal device according to an embodiment of the present application. The terminal device 40 may be a computer device. The terminal device 40 in the present application may include one or more of the following components: a processor 42, a memory 44, and one or more applications, wherein the one or more applications may be stored in the memory 44 and configured to be executed by the one or more processors 42, the one or more applications configured to perform the methods described above as being applied to the advertisement recommendation method embodiments.

Processor 42 may include one or more processing cores. The processor 42 utilizes various interfaces and lines to connect various portions of the overall terminal device 40, perform various functions of the terminal device 40 and process data by executing or executing instructions, programs, code sets, or instruction sets stored in the memory 44, and invoking data stored in the memory 44. Alternatively, the processor 42 may be implemented in hardware in at least one of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), programmable logic array (Programmable Logic Array, PLA). The processor 42 may integrate one or a combination of several of a central processing unit (Central Processing Unit, CPU), a graphics processor (Graphics Processing Unit, GPU), and a modem, etc. The CPU mainly processes an operating system, a user interface, an application program and the like; the GPU is used for being responsible for rendering and drawing of display content; the modem is used to handle wireless communications. It will be appreciated that the modem may not be integrated into the processor 42 and may be implemented solely by a single communication chip.

The Memory 44 may include a random access Memory (Random Access Memory, RAM) or a Read-Only Memory (Read-Only Memory). Memory 44 may be used to store instructions, programs, code, sets of codes, or instruction sets. The memory 44 may include a stored program area and a stored data area, wherein the stored program area may store instructions for implementing an operating system, instructions for implementing at least one function (e.g., a touch function, a sound playing function, an image playing function, etc.), instructions for implementing various method embodiments described below, and the like. The storage data area may also store data created by the terminal device 40 in use, etc.

It will be appreciated by those skilled in the art that the structure shown in fig. 4 is merely a block diagram of a portion of the structure associated with the present application and is not limiting of the terminal device to which the present application is applied, and that a particular terminal device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

In summary, the terminal device provided in the embodiment of the present application is configured to implement the corresponding advertisement recommendation method in the foregoing method embodiment, and has the beneficial effects of the corresponding method embodiment, which is not described herein again.

Referring to fig. 5, a block diagram of a computer readable storage medium according to an embodiment of the present application is shown. The computer readable storage medium 50 has stored therein program code that is executable by a processor to perform the methods described in the advertisement recommendation method embodiments described above.

The computer readable storage medium 50 may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read only memory), an EPROM, a hard disk, or a ROM. Optionally, the computer readable storage medium 50 comprises a non-transitory computer readable medium (non-transitory computer-readable storage medium). The computer readable storage medium 50 has storage space for program code 52 that performs any of the method steps described above. The program code can be read from or written to one or more computer program products. Program code 52 may be compressed, for example, in a suitable form.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An advertisement recommendation method, the method comprising:

2. The method of claim 1, wherein selecting a preset number of advertisement recommendations from the initial set of advertisements to a user based on each click-through rate score comprises:

selecting advertisements with click rate scores greater than a preset threshold value from the initial advertisement set to form a candidate advertisement set;

and selecting a preset number of advertisement recommendations from the candidate advertisement sets to the user.

3. The method of claim 1, wherein training the LightGBM model comprises:

acquiring advertisement click and exposure behavior data samples, and dividing the advertisement click and exposure behavior data samples into a training set and a testing set;

inputting advertisement click and exposure behavior data samples in the training set into the LightGBM model, and adjusting parameters of the LightGBM model until convergence to obtain an initial LightGBM model;

and testing the initial lightGBM model by adopting advertisement click and exposure behavior data samples in the test set, evaluating a feature_importance feature tree of the initial lightGBM model by adopting feature data, and obtaining the trained lightGBM model when the test and the evaluation pass.

4. A method according to claim 3, wherein said adjusting parameters of the LightGBM model comprises:

an LGBMClassifier function is used to adjust the parameters of the LightGBM model.

5. The method of any one of claims 1-4, wherein recalling advertisements according to a preset advertisement recall logic rule comprises:

the advertisement is recalled according to the elastic search index recall rule, and according to the integrated scoring rule of advertisement name, advertisement creator name, and advertisement query sales.

6. The method of any of claims 1-4, wherein the obtaining advertisement click and exposure behavior data of a user in need of advertisement recommendation comprises:

and acquiring advertisement click and exposure behavior data of a user needing advertisement recommendation from a log system by adopting a buried point mode.

7. A method according to claim 3, wherein the feature data comprises user-side feature data, advertisement-side feature data, and user-and advertisement-crossing feature data.

8. An advertisement recommendation device, the device comprising:

9. A terminal device, comprising:

a memory; one or more processors coupled with the memory; one or more applications, wherein the one or more applications are stored in memory and configured to be executed by the one or more processors, the one or more applications configured to perform the method of any of claims 1-7.

10. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a program code, which is callable by a processor for executing the method according to any one of claims 1-7.