CN113222647A - Advertisement recommendation method, system and storage medium based on click rate estimation model - Google Patents
Advertisement recommendation method, system and storage medium based on click rate estimation model Download PDFInfo
- Publication number
- CN113222647A CN113222647A CN202110456025.2A CN202110456025A CN113222647A CN 113222647 A CN113222647 A CN 113222647A CN 202110456025 A CN202110456025 A CN 202110456025A CN 113222647 A CN113222647 A CN 113222647A
- Authority
- CN
- China
- Prior art keywords
- model
- click
- advertisement
- data
- user
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000003860 storage Methods 0.000 title claims abstract description 17
- 238000012549 training Methods 0.000 claims abstract description 35
- 238000012545 processing Methods 0.000 claims abstract description 14
- 238000005457 optimization Methods 0.000 claims abstract description 13
- 238000002474 experimental method Methods 0.000 claims abstract description 11
- 238000004140 cleaning Methods 0.000 claims abstract description 7
- 238000013480 data collection Methods 0.000 claims abstract description 6
- 230000000694 effects Effects 0.000 claims abstract description 6
- 239000000463 material Substances 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000013461 design Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0242—Determining effectiveness of advertisements
- G06Q30/0244—Optimization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0277—Online advertisement
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The advertisement recommendation method, the system and the storage medium based on the click-through rate estimation model, wherein the recommendation method comprises the following steps: data collection, namely cleaning service data to construct a data set; performing characteristic engineering processing on the data set to generate a training sample for model training; performing model training by using Wide & Deep algorithm; adjusting the structural parameters of the model, determining the best super-parameter combination, and optimizing the model; performing an A/B experiment on the optimization model and the online model, and replacing the click rate improved model with the old model; updating the characteristic data to a cloud storage OSS in real time, generating a GRPC interface by using Tensorflow Serving, estimating the click rate of the candidate advertisement list in batch, taking the advertisement with the highest click rate, and recommending exposure, wherein the click rate is more than 0.5. The method and the device have good effect on advertisement recommendation, improve the accuracy of click rate estimation, and are beneficial to improving the recall rate of lost users.
Description
Technical Field
The invention belongs to the field of advertisement recommendation, and particularly relates to an advertisement recommendation method, system and storage medium based on a click rate estimation model.
Background
Click rate estimation is one of the core bases of DSP flow distribution, and the accuracy of click rate estimation plays an important role in recommending online advertisements. In recent years, the click rate estimation model has been changed from the traditional machine learning model to the Deep learning model, and common click rate estimation models include LR, FM, Wide & Deep, Deep FM, and the like. Due to the natural nature of the advertising data: the data volume is large, the dimensionality is high, and the data is sparse, so that the click rate estimation of the advertisement has certain challenge.
Data and features determine the upper limit of machine learning, and models and algorithms only approach this upper limit indefinitely. And selecting a proper algorithm model by combining the service data characteristics, and performing characteristic engineering innovation processing, network structure design and Embedding method tuning are keys for landing a good model. The estimation accuracy of the existing model is still to be improved.
Disclosure of Invention
The invention aims to provide an advertisement recommendation method, system and storage medium based on a click rate estimation model aiming at the problem of low estimation accuracy of a model of advertisement data in the prior art, which can achieve a good effect on advertisement recommendation, improve the estimation accuracy of the click rate and simultaneously contribute to improving the recall rate of lost users.
In order to achieve the purpose, the invention has the following technical scheme:
an advertisement recommendation method based on a click-through rate estimation model comprises the following steps:
-data collection, cleaning of business data, construction of data sets;
-performing feature engineering on the data set to generate training samples for model training;
-model training is performed by using Wide & Deep algorithm based on the Tensorflow framework;
adjusting the model structure parameters, determining the best hyper-parameter combination, and performing model optimization;
performing an A/B experiment on the optimization model and the online model, and replacing the click rate improved model with the old model;
updating the characteristic data to a cloud storage OSS in real time, generating a GRPC interface by using Tensflow Serving, estimating the click rate of a candidate advertisement list in batch, taking the advertisement with the highest click rate, and recommending exposure, wherein the click rate is more than 0.5.
In one embodiment of the invention, the objects of data collection include users, advertisements, request context, media; wherein the user characteristic data comprises: user ID, user commodity preference, user label, user historical funnel depth, recent 3/7/14/28 days exposure times, click times, browsing times, purchase adding times, purchase times and click rate; the advertisement characteristic data includes: advertisement ID, advertisement category, advertisement material ID, advertisement template ID, exposure times, click times and click rate; requesting context feature data includes: time, country, city, channel, media, format size, equipment type, presentation type; the media characteristic data includes: category, rank, PV, UV, click rate.
In an embodiment of the present invention, the specific steps of performing feature engineering processing on a data set are as follows:
step 1), sampling a sample, and adding a positive and negative sample punishment weight;
step 2) dividing data in the data set into continuous features and discrete features to perform normalization and missing value processing;
and 3) carrying out characteristic cross combination.
In an embodiment of the present invention, the continuous features in the user feature data include exposure times, click times, browsing times, and click rate, the continuous features in the advertisement feature data include exposure times, click times, and click rate, the continuous features in the media feature data include ranking, PV, UV, and historical click rate, and the continuous features are normalized by a logarithmic function; and meanwhile, performing equal-frequency bucket discretization on part of continuous features including ranking in the media feature data and purchase times of the user feature data.
The discrete features comprise user ID, user commodity preference, user labels, user historical funnel depth, advertisement ID, advertisement category, advertisement material ID, advertisement template ID, time, country, city, channel, media, format size, equipment type, display type and category, and are subjected to One-Hot coding in a Hash mode.
In an embodiment of the present invention, the missing value processing specifically includes: carrying out feature distribution statistics, analyzing the missing proportion, and giving up the feature when the missing proportion reaches more than 80%; default padding is used for discrete features and mean padding is used for continuous features.
In an embodiment of the present invention, when the step 3) performs the feature cross-combination, the user features and the advertisement and media features are crossed, including the user commodity preferences, the user tags, the user historical funnel depth, and the advertisement ID, the advertisement material ID, the advertisement category and the category are cross-combined.
In an embodiment of the present invention, the concrete steps of training the model by using Wide & Deep algorithm are as follows: inputting discrete features and cross combination features on the Wide layer, inputting continuous features on the Deep layer, and taking a historical data set as a training set and a data set of the latest 1 day as a test set;
performing model tuning, adding Dropout and L2 regularization to prevent overfitting of the model, and introducing Batch Normalization to accelerate convergence of the model; comparing different learners, and selecting Adm with a better effect; and simultaneously trying to optimize different Learning Rate, Batch Size and Embedding parameters, training a model, and finally generating a model file in a SavedModel format.
In an embodiment of the present invention, the specific steps of using the tensoflow Serving to generate the GRPC interface mode include: step 1) starting a Docker, and pulling a Tensorflow Servin Docker mirror image; step 2), generating a prediction interface; step 3), deploying on line, configuring a uniform domain name Tfserve, and generating a final scoring interface service; and 4) performing online reasoning, constructing request data in a batch mode, calling a prediction interface, and finally returning the prediction results of 50 advertisements within 10 ms.
The invention also provides an advertisement recommendation system based on the click-through rate estimation model, which comprises the following steps:
the data set construction module is used for collecting data, cleaning service data and constructing a data set;
the training sample generation module is used for carrying out characteristic engineering processing on the data set to generate a training sample;
the model training module is used for carrying out model training by adopting a Wide & Deep algorithm based on a Tensorflow framework;
the model optimization module is used for adjusting the structural parameters of the model, determining the best super-parameter combination and optimizing the model; performing an A/B experiment on the optimization model and the online model, and replacing the old model with the click rate improved model;
and the recommending module is used for updating the characteristic data to the cloud storage OSS in real time, generating a GRPC interface by using Tensflow Serving, estimating the click rate of the candidate advertisement list in batches, and recommending exposure according to the click rate.
The invention also provides a computer readable storage medium, which stores a computer program, and the computer program, when executed by a processor, implements the steps of the advertisement recommendation method based on the click-through rate estimation model.
Compared with the prior art, the invention has the following beneficial effects:
the method has a good effect on advertisement recommendation, manually designed abstract characteristics such as user commodity preference, user labels, user historical behavior funnel depth and the like, user preference characteristics and advertisement characteristics are combined in a cross mode, and the model can have good memory capacity; meanwhile, a Deep model is adopted, and low-dimensional dense feature input is used through an Embedding method, so that different dimensions of feature vectors are fully crossed, the generalization capability of the model is enhanced, and the recall rate of the model is improved. Meanwhile, the characteristic engineering part is used for mining and analyzing the combined characteristics based on a large amount of detailed business data, so that the characteristics have stronger flexibility, and the interpretability of the model is enhanced. The experiment was evaluated by off-line AUC and on-line ABTest. The AUC of the Wide & Deep model exceeds the original LR model. In the AB Test on-line experiment, the click rate of the Wide & Deep model is improved by 27 percent compared with that of the LR model.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments are briefly described below, it should be understood that the following drawings only show some embodiments of the present invention, and it is obvious for those skilled in the art that other related drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flowchart of an advertisement recommendation method based on a click-through rate estimation model according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, those skilled in the art can also obtain other embodiments without creative efforts.
Referring to fig. 1, an advertisement recommendation method based on a click-through rate estimation model includes the following steps:
s1, collecting data, cleaning service data and constructing a data set;
the objects of data collection include users, advertisements, request context, media; wherein the user characteristic data comprises: user ID, user commodity preference, user label, user historical funnel depth, recent 3/7/14/28 days exposure times, click times, browsing times, purchase adding times, purchase times and click rate; the advertisement characteristic data includes: advertisement ID, advertisement category, advertisement material ID, advertisement template ID, exposure times, click times and click rate; requesting context feature data includes: time, country, city, channel, media, format size, equipment type, presentation type; the media characteristic data includes: category, rank, PV (Page View, i.e. the amount of browsing or clicking on a Page), UV (uniform viewer, which refers to a natural person who accesses and browses the web Page through the internet, and a computer client accessing the website is a Visitor), and click rate.
S2, performing characteristic engineering processing on the data set to generate a training sample for model training;
the specific steps of the characteristic engineering treatment are as follows:
step 1) sampling samples, and adding penalty weights of positive and negative samples.
And 2) dividing the data in the data set into continuous features and discrete features, and performing normalization and missing value processing.
Continuous features:
continuous features in the user feature data comprise exposure times, click times, browsing times and click rates, continuous features in the advertisement feature data comprise exposure times, click times and click rates, continuous features in the media feature data comprise ranking, PV, UV and historical click rates, and the continuous features are normalized by adopting a logarithmic function; and meanwhile, performing equal-frequency bucket discretization on part of continuous features including ranking in the media feature data and purchase times of the user feature data.
The discrete features comprise user ID, user commodity preference, user labels, user historical funnel depth, advertisement ID, advertisement category, advertisement material ID, advertisement template ID, time, country, city, channel, media, format size, equipment type, display type and category, and are subjected to One-Hot coding in a Hash mode.
The missing value processing specifically includes: carrying out feature distribution statistics, analyzing the missing proportion, and giving up the feature when the missing proportion reaches more than 80%; default padding is used for discrete features and mean padding is used for continuous features.
And 3) carrying out characteristic cross combination.
And manually performing characteristic cross combination, and crossing the user characteristics and the advertisement and media characteristics, wherein the step of cross combination comprises the step of cross combination of the commodity preference of a user, the label of the user, the historical funnel depth of the user, the advertisement ID, the advertisement material ID, the advertisement category and the category.
S3, based on a Tensorflow framework, performing model training by using a Wide & Deep algorithm;
the method comprises the following specific steps:
and finally generating a training set through sample construction and feature engineering, and taking the training set as a Wide & Deep training sample. The Wide layer inputs discrete characteristics and cross combination characteristics, including characteristics of cross combination of user ID, commodity preference, label, historical funnel depth, advertisement ID, advertisement category, advertisement material ID, advertisement template ID, request time, country, city, channel, media, layout size, equipment type, display type, media category and user commodity preference, user label, historical funnel depth and advertisement ID, advertisement material ID, advertisement category and media category. The Deep layer input is mainly continuous characteristics besides discrete characteristics, and comprises the exposure times, click times, browsing times and click rates of users, the exposure times, click times and click rates of advertisements, the ranking of media, PV, UV and historical click rates. The historical 60-day data set was used as the training set, and the last 1 day was used as the test set. And (4) optimizing the model, adding Dropout and L2 regularization to prevent the model from being over-fitted, and introducing Batch Normalization to accelerate the convergence of the model. And comparing different learners, and selecting the Adm with better effect. Meanwhile, the optimization of different Learning Rate, Batch Size and Embedding parameters is tried. And training the model, and finally generating a model file in a SavedModel format.
S4, adjusting the model structure parameters, determining the best super parameter combination, and optimizing the model;
s5, performing an A/B experiment on the optimization model and the online model, and replacing the click rate improved model with the old model;
s6, updating the feature data to OSS (Object Storage Service, OSS for short, which is a massive, safe and highly reliable cloud Storage Service provided by Aliskiu cloud), generating GRPC (GRPC is one of RPC frames, a high-performance, open-source and universal RPC frame facing to a server and a mobile terminal and designed based on HTTP/2) by using Tensoflow Serving (a flexible and high-performance application system suitable for a machine learning model and designed for a production environment, and new algorithms and experiments can be easily deployed by means of Tensoflow Serving), and performing click rate prediction on the candidate advertisement list in batches, taking the advertisement with the highest click rate, wherein the click rate is more than 0.5, and recommending and exposing the advertisement.
The specific steps for generating the GRPC interface by using Tensorflow Serving are as follows:
step 1) starting a Docker, and pulling a Tensorflow Serving Docker mirror image; step 2), generating a prediction interface; step 3), deploying on line, configuring a uniform domain name Tfserve, and generating a final scoring interface service; and 4) performing online reasoning, constructing request data in a batch mode, calling a prediction interface, and finally returning the prediction results of 50 advertisements within 10 ms.
The click rate comparison results of the Wide & Deep model and the LR model used in the invention are shown in the following table:
in the ABTest online experiment, the click rate of the Wide & Deep model is improved by 27 percent compared with that of the LR model.
An advertisement recommendation system based on a click-through rate estimation model comprises:
the data set construction module is used for collecting data, cleaning service data and constructing a data set;
the training sample generation module is used for carrying out characteristic engineering processing on the data set to generate a training sample;
the model training module is used for carrying out model training by adopting a Wide & Deep algorithm based on a Tensorflow framework;
the model optimization module is used for adjusting the structural parameters of the model, determining the best super-parameter combination and optimizing the model; performing an A/B experiment on the optimization model and the online model, and replacing the old model with the click rate improved model;
and the recommending module is used for updating the characteristic data to the cloud storage OSS in real time, generating a GRPC interface by using Tensflow Serving, estimating the click rate of the candidate advertisement list in batches, and recommending exposure according to the click rate.
A computer-readable storage medium, storing a computer program which, when executed by a processor, implements the steps of the click-through rate estimation model-based advertisement recommendation method.
The computer program may be partitioned into one or more modules/units stored in the memory and executed by the processor to perform a click-through rate prediction model based advertisement recommendation method.
The memory may be used to store computer programs and/or modules that the processor performs various functions of the system by executing or otherwise executing the computer programs and/or modules stored in the memory and invoking data stored in the memory.
The advertisement recommendation method, the advertisement recommendation system and the storage medium based on the click rate estimation model have the following advantages that:
1. the click rate estimation accuracy is high, feature design and feature combination are manually carried out based on understanding of business data, displayed, direct and associated rules and other features are used as Wide part input models, and the memory capacity of the models is improved. 2. The recall rate is high, namely a certain generalization ability is achieved. And the Embedding tuning is carried out on different ID characteristics and the network design is used as a Deep part, so that the model can capture more information of nonlinear characteristics and combination characteristics, and the expression capability is enhanced.
The above-mentioned embodiments are only preferred embodiments of the present invention, and are not intended to limit the technical solution of the present invention, and it should be understood by those skilled in the art that the technical solution can be modified and replaced by a plurality of simple modifications and replacements without departing from the spirit and principle of the present invention, and the modifications and replacements also fall into the protection scope covered by the claims.
Claims (10)
1. An advertisement recommendation method based on a click-through rate estimation model is characterized by comprising the following steps:
-data collection, cleaning of business data, construction of data sets;
-performing feature engineering on the data set to generate training samples for model training;
-model training is performed by using Wide & Deep algorithm based on the Tensorflow framework;
adjusting the model structure parameters, determining the best hyper-parameter combination, and performing model optimization;
performing an A/B experiment on the optimization model and the online model, and replacing the click rate improved model with the old model;
updating the characteristic data to a cloud storage OSS in real time, generating a GRPC interface by using Tensflow Serving, estimating the click rate of a candidate advertisement list in batch, taking the advertisement with the highest click rate, and recommending exposure, wherein the click rate is more than 0.5.
2. The advertisement recommendation method based on click-through rate estimation model as claimed in claim 1, wherein: the objects of data collection include users, advertisements, request context, media; wherein the user characteristic data comprises: user ID, user commodity preference, user label, user historical funnel depth, recent 3/7/14/28 days exposure times, click times, browsing times, purchase adding times, purchase times and click rate; the advertisement characteristic data includes: advertisement ID, advertisement category, advertisement material ID, advertisement template ID, exposure times, click times and click rate; requesting context feature data includes: time, country, city, channel, media, format size, equipment type, presentation type; the media characteristic data includes: category, rank, PV, UV, click rate.
3. The advertisement recommendation method based on click-through rate estimation model as claimed in claim 2, wherein the specific steps of performing feature engineering processing on the data set are as follows:
step 1), sampling a sample, and adding a positive and negative sample punishment weight;
step 2) dividing data in the data set into continuous features and discrete features to perform normalization and missing value processing;
and 3) carrying out characteristic cross combination.
4. The advertisement recommendation method based on click-through rate estimation model as claimed in claim 3, wherein:
continuous features in the user feature data comprise exposure times, click times, browsing times and click rates, continuous features in the advertisement feature data comprise exposure times, click times and click rates, continuous features in the media feature data comprise ranking, PV, UV and historical click rates, and the continuous features are normalized by adopting a logarithmic function; and meanwhile, performing equal-frequency bucket discretization on part of continuous features including ranking in the media feature data and purchase times of the user feature data.
The discrete features comprise user ID, user commodity preference, user labels, user historical funnel depth, advertisement ID, advertisement category, advertisement material ID, advertisement template ID, time, country, city, channel, media, format size, equipment type, display type and category, and are subjected to One-Hot coding in a Hash mode.
5. The advertisement recommendation method based on click-through rate estimation model as claimed in claim 4, wherein the missing value processing specifically comprises: carrying out feature distribution statistics, analyzing the missing proportion, and giving up the feature when the missing proportion reaches more than 80%; default padding is used for discrete features and mean padding is used for continuous features.
6. The advertisement recommendation method based on click-through rate estimation model as claimed in claim 4, wherein in the step 3) of cross-combining features, the user features and the advertisement and media features are cross-combined, including cross-combining the user commodity preferences, the user tags, the user historical funnel depth and the advertisement ID, the advertisement material ID, the advertisement category and the category.
7. The advertisement recommendation method based on the click-through rate estimation model as claimed in claim 6, wherein the specific steps of model training by using Wide & Deep algorithm are as follows: inputting discrete features and cross combination features on the Wide layer, inputting continuous features on the Deep layer, and taking a historical data set as a training set and a data set of the latest 1 day as a test set;
performing model tuning, adding Dropout and L2 regularization to prevent overfitting of the model, and introducing Batch Normalization to accelerate convergence of the model; comparing different learners, and selecting Adm with a better effect; and simultaneously trying to optimize different Learning Rate, Batch Size and Embedding parameters, training a model, and finally generating a model file in a SavedModel format.
8. The advertisement recommendation method based on click-through rate estimation model as claimed in claim 1, wherein the specific step of generating a GRPC interface using the tensrflow Serving comprises:
step 1) starting a Docker, and pulling a Tensorflow Serving Docker mirror image; step 2), generating a prediction interface; step 3), deploying on line, configuring a uniform domain name Tfserve, and generating a final scoring interface service; and 4) performing online reasoning, constructing request data in a batch mode, calling a prediction interface, and finally returning the prediction results of 50 advertisements within 10 ms.
9. An advertisement recommendation system based on a click-through rate estimation model is characterized by comprising:
the data set construction module is used for collecting data, cleaning service data and constructing a data set;
the training sample generation module is used for carrying out characteristic engineering processing on the data set to generate a training sample;
the model training module is used for carrying out model training by adopting a Wide & Deep algorithm based on a Tensorflow framework;
the model optimization module is used for adjusting the structural parameters of the model, determining the best super-parameter combination and optimizing the model; performing an A/B experiment on the optimization model and the online model, and replacing the old model with the click rate improved model;
and the recommending module is used for updating the characteristic data to the cloud storage OSS in real time, generating a GRPC interface by using Tensflow Serving, estimating the click rate of the candidate advertisement list in batches, and recommending exposure according to the click rate.
10. A computer-readable storage medium storing a computer program, characterized in that: the computer program when being executed by a processor implements the steps of the method for recommending advertisements based on a click-through rate estimation model according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110456025.2A CN113222647B (en) | 2021-04-26 | 2021-04-26 | Advertisement recommendation method, system and storage medium based on click rate estimation model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110456025.2A CN113222647B (en) | 2021-04-26 | 2021-04-26 | Advertisement recommendation method, system and storage medium based on click rate estimation model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113222647A true CN113222647A (en) | 2021-08-06 |
| CN113222647B CN113222647B (en) | 2023-11-28 |
Family
ID=77089472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110456025.2A Active CN113222647B (en) | 2021-04-26 | 2021-04-26 | Advertisement recommendation method, system and storage medium based on click rate estimation model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113222647B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113704615A (en) * | 2021-08-30 | 2021-11-26 | 万汇互联(深圳)科技有限公司 | Deep interest network recommendation method based on multiple modes |
| CN113837483A (en) * | 2021-09-29 | 2021-12-24 | 深圳市易平方网络科技有限公司 | Advertisement flow pre-estimation processing method and device based on wireless receiving device and terminal |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20110063160A (en) * | 2009-12-04 | 2011-06-10 | 동국대학교 산학협력단 | Apparatus and method for providing advertisement |
| CN105590240A (en) * | 2015-12-30 | 2016-05-18 | 合一网络技术(北京)有限公司 | Discrete calculating method of brand advertisement effect optimization |
| CN108228824A (en) * | 2017-12-29 | 2018-06-29 | 暴风集团股份有限公司 | Recommendation method, apparatus, electronic equipment, medium and the program of a kind of video |
| CN108416625A (en) * | 2018-02-28 | 2018-08-17 | 阿里巴巴集团控股有限公司 | The recommendation method and apparatus of marketing product |
| CN108596645A (en) * | 2018-03-13 | 2018-09-28 | 阿里巴巴集团控股有限公司 | A kind of method, apparatus and equipment of information recommendation |
| WO2018212710A1 (en) * | 2017-05-19 | 2018-11-22 | National University Of Singapore | Predictive analysis methods and systems |
| US20190325293A1 (en) * | 2018-04-19 | 2019-10-24 | National University Of Singapore | Tree enhanced embedding model predictive analysis methods and systems |
| CN110619540A (en) * | 2019-08-13 | 2019-12-27 | 浙江工业大学 | Click stream estimation method of neural network |
| CN110728541A (en) * | 2019-10-11 | 2020-01-24 | 广州市丰申网络科技有限公司 | Information stream media advertisement creative recommendation method and device |
| CN110852793A (en) * | 2019-10-28 | 2020-02-28 | 北京深演智能科技股份有限公司 | Document recommendation method and device and electronic equipment |
| CN111435507A (en) * | 2019-01-11 | 2020-07-21 | 腾讯科技（北京）有限公司 | Advertisement anti-cheating method and device, electronic equipment and readable storage medium |
| CN112434184A (en) * | 2020-12-15 | 2021-03-02 | 四川长虹电器股份有限公司 | Deep interest network sequencing method based on historical movie posters |
-
2021
- 2021-04-26 CN CN202110456025.2A patent/CN113222647B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20110063160A (en) * | 2009-12-04 | 2011-06-10 | 동국대학교 산학협력단 | Apparatus and method for providing advertisement |
| CN105590240A (en) * | 2015-12-30 | 2016-05-18 | 合一网络技术(北京)有限公司 | Discrete calculating method of brand advertisement effect optimization |
| WO2018212710A1 (en) * | 2017-05-19 | 2018-11-22 | National University Of Singapore | Predictive analysis methods and systems |
| CN108228824A (en) * | 2017-12-29 | 2018-06-29 | 暴风集团股份有限公司 | Recommendation method, apparatus, electronic equipment, medium and the program of a kind of video |
| CN108416625A (en) * | 2018-02-28 | 2018-08-17 | 阿里巴巴集团控股有限公司 | The recommendation method and apparatus of marketing product |
| CN108596645A (en) * | 2018-03-13 | 2018-09-28 | 阿里巴巴集团控股有限公司 | A kind of method, apparatus and equipment of information recommendation |
| US20190325293A1 (en) * | 2018-04-19 | 2019-10-24 | National University Of Singapore | Tree enhanced embedding model predictive analysis methods and systems |
| CN111435507A (en) * | 2019-01-11 | 2020-07-21 | 腾讯科技（北京）有限公司 | Advertisement anti-cheating method and device, electronic equipment and readable storage medium |
| CN110619540A (en) * | 2019-08-13 | 2019-12-27 | 浙江工业大学 | Click stream estimation method of neural network |
| CN110728541A (en) * | 2019-10-11 | 2020-01-24 | 广州市丰申网络科技有限公司 | Information stream media advertisement creative recommendation method and device |
| CN110852793A (en) * | 2019-10-28 | 2020-02-28 | 北京深演智能科技股份有限公司 | Document recommendation method and device and electronic equipment |
| CN112434184A (en) * | 2020-12-15 | 2021-03-02 | 四川长虹电器股份有限公司 | Deep interest network sequencing method based on historical movie posters |
Non-Patent Citations (1)
| Title |
|---|
| 林启迪: "基于宽深度模型的广告点击率预估方法", 华南理工大学硕士学位论文, pages 1 - 68 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113704615A (en) * | 2021-08-30 | 2021-11-26 | 万汇互联(深圳)科技有限公司 | Deep interest network recommendation method based on multiple modes |
| CN113837483A (en) * | 2021-09-29 | 2021-12-24 | 深圳市易平方网络科技有限公司 | Advertisement flow pre-estimation processing method and device based on wireless receiving device and terminal |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113222647B (en) | 2023-11-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105468742B (en) | The recognition methods of malice order and device | |
| EP4181026A1 (en) | Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium | |
| JP5454357B2 (en) | Information processing apparatus and method, and program | |
| CN110162693A (en) | A kind of method and server of information recommendation | |
| JP6854748B2 (en) | Information providing equipment, information providing method, and program | |
| CN112765480B (en) | Information pushing method and device and computer readable storage medium | |
| US20130144854A1 (en) | Modeling actions for entity-centric search | |
| JP5507607B2 (en) | Content providing apparatus, low rank approximate matrix generating apparatus, content providing method, low rank approximate matrix generating method, and program | |
| CN107122455A (en) | A kind of network user's enhancing method for expressing based on microblogging | |
| US11651255B2 (en) | Method and apparatus for object preference prediction, and computer readable medium | |
| CN113569129A (en) | Click rate prediction model processing method, content recommendation method, device and equipment | |
| CN106997347A (en) | Information recommendation method and server | |
| CN108632311A (en) | Information-pushing method and device | |
| CN113222647B (en) | Advertisement recommendation method, system and storage medium based on click rate estimation model | |
| JP7498248B2 (en) | Content recommendation and sorting model training method, device, equipment, storage medium, and computer program | |
| CN110008397A (en) | A kind of recommended models training method and device | |
| CN113254679A (en) | Multimedia resource recommendation method and device, electronic equipment and storage medium | |
| CN109690581A (en) | User guided system and method | |
| CN108833933A (en) | A kind of method and system for recommending video flow using support vector machines | |
| CN112633931A (en) | Click rate prediction method, device, electronic equipment and medium | |
| CN113469752A (en) | Content recommendation method and device, storage medium and electronic equipment | |
| CN111429214B (en) | Transaction data-based buyer and seller matching method and device | |
| Jiang et al. | Retweet prediction using social-aware probabilistic matrix factorization | |
| CN113111198B (en) | Demonstration manuscript recommendation method based on collaborative filtering algorithm and related equipment | |
| CN116385076A (en) | Advertisement recommendation method, advertisement recommendation device, terminal equipment and computer readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |