IP Library › Granted Patent US 12,236,457
Granted Patent B2
US 12,236,457 · App. 17/880,605 · Granted Feb 25, 2025

Systems and methods for multidimensional knowledge transfer for click through rate prediction

Inventors: Qijun Zhu (Hong Kong, HK); Xin Zheng (Hong Kong, HK); Ming Ming Tan (Augusta, GA)
Assignee: Hong Kong Applied Science and Technology Research Institute Company Limited
G06Q30/0272
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,236,457
App. No.
17/880,605
Granted
Feb 25, 2025
Kind
B2
Abstract

A multidimensional knowledge transfer model for predicting CTR of a computational ad, the multidimensional knowledge transfer model comprises: a pre-processor for building an ad group node graph based on similarities among ad group nodes, an ad campaign node graph from merging the ad group node graph, and an ad account node graph from merging the ad campaign node graph. The multidimensional knowledge transfer model further comprises a multi-knowledge CTR prediction model for each of the ad account, ad campaign, and ad group layers. The multi-knowledge CTR prediction model predicts the respective node's CTR from the ad account node graph, ad campaign node graph, or ad group node graph, features of the audience group, and features of the node having its CTR predicted appended with the hidden vector of its parent node extracted from the upper layer multi-knowledge CTR prediction model.

Claims (74)

1. A system for predicting a probability of a computational advertisement (ad) displayed on a website or an online electronic user interface will be accessed when shown to an audience group, the system comprising:

a multidimensional knowledge transfer model, implemented by one or more processors, the multidimensional knowledge transfer model comprising:

a logical pre-processor configured to build:

an ad group node graph of a plurality of ad group nodes based on one or more feature similarities among the ad group nodes;

an ad campaign node graph of one or more ad campaign nodes from merging nodes in the ad group node graph of ad group nodes belonging to each of the ad campaign nodes; and

an ad account node graph of one or more ad account nodes from merging nodes in the ad campaign node graph of ad campaign nodes belonging to each of the ad account nodes;

an ad account multi-knowledge click-through-rate (CTR) prediction model comprising a first horizontal knowledge transfer model for solving cold start problem and a first hierarchical knowledge transfer model for solving imbalanced data problem, the ad account multi-knowledge CTR prediction model having trained to predict an ad account CTR for an ad account having its CTR predicted from the ad account node graph, features of the audience group, features of an ad account node of the ad account having its CTR predicted by the first horizontal knowledge transfer model, and features of other ad account nodes;

an ad campaign multi-knowledge CTR prediction model comprising a second horizontal knowledge transfer model for solving cold start problem and a second hierarchical knowledge transfer model for solving imbalanced data problem, the ad campaign multi-knowledge CTR prediction model having trained to predict an ad campaign CTR for an ad campaign having its CTR predicted from the ad campaign node graph, features of the audience group, features of an ad campaign node of the ad campaign having its CTR predicted appended with an ad account node hidden vector, and features of other ad campaign nodes;

wherein the ad campaign having its CTR predicted belongs to the ad account having its CTR predicted; and

wherein the ad account node hidden vector is extracted from the ad account multi-knowledge CTR prediction model in predicting the CTR of the ad account having its CTR predicted; and

an ad group multi-knowledge prediction model comprising a third horizontal knowledge transfer model for solving cold start problem and a third hierarchical knowledge transfer model for solving imbalanced data problem, the ad group multi-knowledge prediction model having trained to predict an ad group CTR for an ad group having its CTR predicted from the ad group node graph, features of the audience group, features of an ad group node of the ad group having its CTR predicted appended with an ad campaign node hidden vector, and features of other ad group nodes;

wherein the ad group having its CTR predicted belongs to the ad campaign having its CTR predicted; and

wherein the ad campaign node hidden vector is extracted from the ad campaign multi-knowledge CTR prediction model in predicting the CTR of the ad campaign having its CTR predicted;

wherein the ad group having its CTR predicted comprises the computational ad, and the predicted ad group CTR indicates the probability of the computational ad will be accessed; and

wherein each of the ad account multi-knowledge CTR prediction model, the ad campaign multi-knowledge CTR prediction model, and the ad group multi-knowledge prediction model is implemented by a Support Vector Machine neural network, a Factorization Machine neural network, a Parallel-Structure deep learning neural network, a Serial-Structure deep learning neural network, or a General Interest-Structure deep learning neural network.

2. The system of claim 1 ,

wherein the logical pre-processor is further configured to build the ad group node graph based on the feature similarities among the ad group nodes comprising one or more keywords of the landing pages associated with one or more ads in each of the ad group nodes.

3. The system of claim 1 ,

wherein each of the first, the second, and the third horizontal knowledge transfer models comprises one or more of a Graph Convolution Network (GCN) and a Regression Artificial Neural Network (ANN).

4. The system of claim 1 ,

wherein during training, the ad account multi-knowledge CTR prediction model is trained first with ad account settings and performance data of a plurality of ad accounts;

wherein the ad account performance data is an aggregate of all ad campaign performance data of all ad campaigns belonging to the ad accounts; and

wherein each of the ad campaign performance data is an aggregate of all ad group performance data of all ad groups belonging to the ad campaigns.

5. The system of claim 4 ,

wherein the ad campaign multi-knowledge CTR prediction model is trained after the training of the ad account multi-knowledge CTR prediction model;

wherein hidden vectors of the ad account nodes are embedded into the ad campaign multi-knowledge CTR prediction model by way of appending each of the hidden vectors of ad account nodes to each of the ad account nodes' children ad campaign node's features being input to the ad campaign multi-knowledge CTR prediction model.

6. The system of claim 5 ,

wherein the ad group multi-knowledge CTR prediction model is trained after the training of the ad campaign multi-knowledge CTR prediction model;

wherein hidden vectors of the ad campaign nodes are embedded into the ad group multi-knowledge CTR prediction model by way of appending each of the hidden vectors of ad campaign nodes to each of the ad campaign nodes' children ad group node's features being input to the ad group multi-knowledge CTR prediction model.

7. The system of claim 1 , wherein during training:

an ad group node graph of one or more ad group nodes, settings and performance features of the ad group nodes, and features of one or more audience groups, and past CTR values of pre-existing ad group nodes are taken as training dataset for training the ad group multi-knowledge CTR prediction model;

an ad campaign node graph of one or more ad campaign nodes, settings and performance features of the ad campaign nodes, and the features of the audience groups, and past CTR values of pre-existing ad campaign nodes are taken as training dataset for training the ad campaign multi-knowledge CTR prediction model;

an ad account node graph of one or more ad account nodes, settings and performance features of the ad account nodes, and the features of the audience groups, and past CTR values of pre-existing ad account nodes are taken as training dataset for training the ad account multi-knowledge CTR prediction model.

8. The system of claim 1 , wherein the features of the ad group node of the ad group having its CTR predicted and the other ad group nodes comprise:

settings features comprising: account ID, campaign ID, ad group ID, ad network type, interaction type, target location, and target language; and

performance features comprising: clicks, impressions, click-through-rate (CTR), cost, interactions, conversions, engagements video views, active view measurability, active view impressions, active view measurable cost, active view measurable impressions, active view viewability, active CTR, active view cost-per-thousand impressions (CPM), average cost-per-click, average CPM, average cost-per-engagement (CPE), and average cost-per-view (CPV).

9. The system of claim 1 , wherein the features of the audience group comprise audience group ID, age, gender, and placement type.

10. A computer-implemented method for predicting a probability of a computational advertisement (ad) displayed on a website or an online electronic user interface will be accessed when shown to an audience group, the method comprising:

executing a multidimensional knowledge transfer model to predict the probability of the computational ad will be accessed, the multidimensional knowledge transfer model comprising:

a logical pre-processor configured to build:

an ad group node graph of a plurality of ad group nodes based on one or more feature similarities among the ad group nodes;

an ad campaign node graph of one or more ad campaign nodes from merging nodes in the ad group node graph of ad group nodes belonging to each of the ad campaign nodes; and

an ad account node graph of one or more ad account nodes from merging nodes in the ad campaign node graph of ad campaign nodes belonging to each of the ad account nodes;

an ad account multi-knowledge click-through-rate (CTR) prediction model comprising a first horizontal knowledge transfer model for solving cold start problem and a first hierarchical knowledge transfer model for solving imbalanced data problem, the ad account multi-knowledge CTR prediction model having trained to predict an ad account CTR for an ad account having its CTR predicted from the ad account node graph, features of the audience group, features of an ad account node of the ad account having its CTR predicted by the first horizontal knowledge transfer model, and features of other ad account nodes;

an ad campaign multi-knowledge CTR prediction model comprising a second horizontal knowledge transfer model for solving cold start problem and a second hierarchical knowledge transfer model for solving imbalanced data problem, the ad campaign multi-knowledge CTR prediction model having trained to predict an ad campaign CTR for an ad campaign having its CTR predicted from the ad campaign node graph, features of the audience group, features of an ad campaign node of the ad campaign having its CTR predicted appended with an ad account node hidden vector, and features of other ad campaign nodes;

wherein the ad campaign having its CTR predicted belongs to the ad account having its CTR predicted; and

wherein the ad account node hidden vector is extracted from the ad account multi-knowledge CTR prediction model in predicting the CTR of the ad account having its CTR predicted; and

an ad group multi-knowledge prediction model comprising a third horizontal knowledge transfer model for solving cold start problem and a third hierarchical knowledge transfer model for solving imbalanced data problem, the ad group multi-knowledge prediction model having trained to predict an ad group CTR for an ad group having its CTR predicted from the ad group node graph, features of the audience group, features of an ad group node of the ad group having its CTR predicted appended with an ad campaign node hidden vector, and features of other ad group nodes;

wherein the ad group having its CTR predicted belongs to the ad campaign having its CTR predicted; and

wherein the ad campaign node hidden vector is extracted from the ad campaign multi-knowledge CTR prediction model in predicting the CTR of the ad campaign having its CTR predicted;

wherein the ad group having its CTR predicted comprises the computational ad, and the predicted ad group CTR indicates the probability of the computational ad will be accessed; and

wherein each of the ad account multi-knowledge CTR prediction model, the ad campaign multi-knowledge CTR prediction model, and the ad group multi-knowledge prediction model is implemented by a Support Vector Machine neural network, a Factorization Machine neural network, a Parallel-Structure deep learning neural network, a Serial-Structure deep learning neural network, or a General Interest-Structure deep learning neural network.

11. The method of claim 10 ,

wherein the logical pre-processor is further configured to build the ad group node graph based on the feature similarities among the ad group nodes comprising one or more keywords of the landing pages associated with one or more ads in each of the ad group nodes.

12. The method of claim 10 ,

wherein each of the first, the second, and the third horizontal knowledge transfer models comprises one or more of a Graph Convolution Network (GCN) and a Regression Artificial Neural Network (ANN).

13. The method of claim 10 ,

wherein during training, the ad account multi-knowledge CTR prediction model is trained first with ad account settings and performance data of a plurality of ad accounts;

wherein the ad account performance data is an aggregate of ad campaign performance data of all ad campaigns belonging to the ad accounts; and

wherein the ad campaign performance data is an aggregate of ad group performance data of all ad groups belonging to the ad campaigns.

14. The method of claim 13 ,

wherein the ad campaign multi-knowledge CTR prediction model is trained after the training of the ad account multi-knowledge CTR prediction model;

wherein hidden vectors of the ad account nodes are embedded into the ad campaign multi-knowledge CTR prediction model by way of appending each of the hidden vectors of ad account nodes to each of the ad account nodes' children ad campaign node's features being input to the ad campaign multi-knowledge CTR prediction model.

15. The method of claim 14 ,

wherein the ad group multi-knowledge CTR prediction model is trained after the training of the ad campaign multi-knowledge CTR prediction model;

wherein hidden vectors of the ad campaign nodes are embedded into the ad group multi-knowledge CTR prediction model by way of appending each of the hidden vectors of ad campaign nodes to each of the ad campaign nodes' children ad group node's features being input to the ad group multi-knowledge CTR prediction model.

16. The method of claim 10 , wherein during training:

an ad group node graph of one or more ad group nodes, settings and performance features of the ad group nodes, features of one or more audience groups, and past CTR values of pre-existing ad group nodes are taken as training dataset for training the ad group multi-knowledge CTR prediction model;

an ad campaign node graph of one or more ad campaign nodes, settings and performance features of the ad campaign nodes, and the features of the audience groups, and past CTR values of pre-existing ad campaign nodes are taken as training dataset for training the ad campaign multi-knowledge CTR prediction model;

an ad account node graph of one or more ad account nodes, settings and performance features of the ad account nodes, and the features of the audience groups, and past CTR values of pre-existing ad account nodes are taken as training dataset for training the ad account multi-knowledge CTR prediction model.

17. The method of claim 10 , wherein the features of the ad group node of the ad group having its CTR predicted and the other ad group nodes comprise:

settings features comprising: account ID, campaign ID, ad group ID, ad network type, interaction type, target location, and target language; and

performance features comprising: clicks, impressions, click-through-rate (CTR), cost, interactions, conversions, engagements video views, active view measurability, active view impressions, active view measurable cost, active view measurable impressions, active view viewability, active CTR, active view cost-per-thousand impressions (CPM), average cost-per-click, average CPM, average cost-per-engagement (CPE), and average cost-per-view (CPV).

18. The method of claim 10 , wherein the features of the audience group comprise audience group ID, age, gender, and placement type.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Aug 4, 2022
From: ZHU, QIJUN; ZHENG, XIN; TAN, MING MING
To: HONG KONG APPLIED SCIENCE AND TECHNOLOGY RESEARCH INSTITUTE COMPANY LIMITED
Reel/Frame 060714/0327 →
Continuity (1)
Related Publication 20240046314A1 · Feb 8, 2024
References Cited (26)
US 9549017B2 · Zhang et al. · 2017 [cited by applicant]
US 10559004B2 · Yang et al. · 2020 [cited by applicant]
US 20140068011A1 · Zhang · 2014 [cited by examiner]
US 20170161773A1 · Xu · 2017 [cited by examiner]
US 20180314944A1 · Li · 2018 [cited by examiner]
US 20200265466A1 · Cheng · 2020 [cited by examiner]
CN 109615060A · 2019 [cited by applicant]
CN 110633421A · 2019 [cited by applicant]
CN 110738314A · 2020 [cited by applicant]
CN 112288471A · 2021 [cited by applicant]
CN 113343095A · 2021 [cited by applicant]
CN 113723538A · 2021 [cited by applicant]
E. Mark Milankovich, “The Cold Start Problem for Recommender Systems”, available on Jul. 14, 2015, retrieved from <<https://medium.com/@markmilankovich/the-cold-start-problem-for-recommender-systems-89a76505a7>> (Year: … [cited by examiner]
Jiang Zilong et al., “A CTR Prediction Approach for Advertising based on Embedding Model and Deep Learning”, 2018 IEEE Intl Conf on Parallel & Distributed Processing with Applications, Ubiquitous Computing & Communicati… [cited by applicant]
International Search Report and Written Opinion of corresponding PCT application No. PCT/CN2022/111732 mailed on Apr. 7, 2023. [cited by applicant]
“Support-vector machine”, https://en.wikipedia.org/wiki/Support-vector_machine. [cited by applicant]
Steffen Rendle, “Factorization machines”, 2010 IEEE International Conference on Data Mining, 2010. [cited by applicant]
Heng-Tze Cheng et al., “Wide & Deep Learning for Recommender Systems”, Proceedings of the 1st Workshop on Deep Learning for Recommender Systems, Sep. 2016. [cited by applicant]
Huifeng Guo et al., “DeepFM: A Factorization-Machine based Neural Network for CTR Prediction”, Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, 2017, pp. 1725-1731. [cited by applicant]
Weinan Zhang et al., “Deep Learning over Multi-field Categorical Data—A Case Study on User Response Prediction”, European Conference on Information Retrieval, 2016, pp. 45-57. [cited by applicant]
Yanru Qu et al., “Product-based Neural Networks for User Response Prediction”, 2016 IEEE 16th International Conference on Data Mining (ICDM), Dec. 2016. [cited by applicant]
Guorui Zhou et al., “Deep Interest Network for Click-Through Rate Prediction”, Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Jul. 2018, pp. 1059-1068. [cited by applicant]
Paul Covington et al., “Deep Neural Networks for YouTube Recommendations”, Proceedings of the 10th ACM Conference on Recommender Systems, Sep. 2016, pp. 191-198. [cited by applicant]
Thomas N. Kipf et al., “Semi-Supervised Classification with Graph Convolutional Networks”, International Conference on Learning Representations (ICLR), Feb. 22, 2017, pp. 1-14. [cited by applicant]
“Artificial neural network”, https://en.wikipedia.org/wiki/Artificial_neural_network. [cited by applicant]
Google Ads API Reports (Metrics), https://developers.google.com/google-ads/api/fields/v10/metrics. [cited by applicant]
Cited By (1)
US 12,613,863