Methods and systems for near real-time lookalike audience expansion in ads targeting

Patent metadata

Field	Value
Publication number	US-10853847-B2
Application number	US-201615154613-A
Country	US
Kind code	B2
Filing date	May 13, 2016
Priority date	May 13, 2016
Publication date	Dec 1, 2020
Grant date	Dec 1, 2020

Abstract

A signature matrix is used to test each online event for inclusion of a user in a pre-generated lookalike audience. Locality Sensitive Hashing is used to compile the signature matrix to dramatically reduce memory storage requirements and processing time for each received online event. A technique of producing incremental signature matrices between generation of the full signature matrix helps to enable near-real time performance in processing each received event to see if a user associated with that event should be added to one of the lookalike audiences pre-generated by advertiser.

First claim

What is claimed is:

1. A system comprising:
a processor; and
memory comprising processor-executable instructions that when executed by the processor cause the processor to perform operations comprising:
defining a lookalike audience using one or more models;
receiving, from users, a stream of events;
receiving, via an application program interface (API), seed list information from a user device;
receiving, via the API, lookalike audience information from one or more devices associated with one or more advertisers;
updating the one or more models based upon at least one of the seed list information or the lookalike audience information;
spawning, via one or more database loader threads, one or more pre-compiler threads;
preparing data, via the one or more pre-compiler threads, by (i) fetching seed data, and (ii) storing the seed data in a user profile cache;
signaling, via the one or more pre-compiler threads, one or more signature matrix compiler threads;
building, based upon the seed data and the signaling and via the one or more signature matrix compiler threads, one or more signature matrices;
processing a first event of the stream of events based upon the one or more signature matrices, wherein the first event is associated with a first user; and
determining a degree of similarity between the first user and one or more members of the lookalike audience based upon the processing.

2. The system of claim 1 , wherein the operations comprise:
receiving user profile information from a user database, wherein the building is based upon a hash function of at least one of the user profile information or information about one or more lookalike audiences.

3. The system of claim 1 , wherein the building comprises:
building a first signature matrix based upon two or more lookalike audiences.

4. The system of claim 3 , wherein the operations comprise:
determining a first result based upon a first lookup to the first signature matrix for the first event;
determining a second result based upon a second lookup to a second signature matrix; and
determining the lookalike audience based upon a combination of the first result and the second result.

5. The system of claim 1 , wherein the building is based upon k functions for N users.

6. The system of claim 3 , wherein the operations comprise:
updating the two or more lookalike audiences.

7. The system of claim 6 , wherein the operations comprise:
building a second signature matrix responsive to the updating the two or more lookalike audiences.

8. The system of claim 6 , wherein the operations comprise:
building a second signature matrix.

9. A method comprising:
defining, by an audience expansion engine, a lookalike audience using one or more models;
receiving, by the audience expansion engine and from users, a stream of events;
receiving, by the audience expansion engine and via an application program interface (API), seed list information from a user device;
receiving, by the audience expansion engine and via the API, lookalike audience information from one or more devices associated with one or more advertisers;
updating, by the audience expansion engine, the one or more models based upon at least one of the seed list information or the lookalike audience information;
spawning, by the audience expansion engine and via one or more database loader threads, one or more pre-compiler threads;
preparing data, by the audience expansion engine and via the one or more pre-compiler threads, by (i) fetching seed data, and (ii) storing the seed data in a user profile cache;
signaling, by the audience expansion engine and via the one or more pre-compiler threads, one or more signature matrix compiler threads;
building, by the audience expansion engine and based upon the seed data and the signaling and via the one or more signature matrix compiler threads, one or more signature matrices;
processing, by the audience expansion engine, a first event of the stream of events based upon the one or more signature matrices, wherein the first event is associated with a first user; and
determining, by the audience expansion engine, a degree of similarity between the first user and one or more members of the lookalike audience based upon the processing.

10. The method of claim 9 , wherein the building comprises:
building, by the audience expansion engine, a first signature matrix based upon two or more lookalike audiences.

11. The method of claim 10 , comprising:
determining, by the audience expansion engine, a first result based upon a first lookup to the first signature matrix for the first event;
determining, by the audience expansion engine, a second result based upon a second lookup to a second signature matrix; and
determining, by the audience expansion engine, the lookalike audience based upon a combination of the first result and the second result.

12. The method of claim 10 , comprising:
updating, by the audience expansion engine, the two or more lookalike audiences.

13. The method of claim 12 , comprising:
building, by the audience expansion engine, a second signature matrix responsive to the updating the two or more lookalike audiences.

14. The method of claim 12 , comprising:
building, by the audience expansion engine, a second signature matrix.

15. A non-transitory computer-readable storage medium comprising instructions that when executed by a processor cause the processor to perform operations comprising:
defining a lookalike audience using one or more models;
receiving, from users, a stream of events;
receiving, via an application program interface (API), seed list information from a user device;
receiving, via the API, lookalike audience information from one or more devices associated with one or more advertisers;
updating the one or more models based upon at least one of the seed list information or the lookalike audience information;
spawning, via one or more database loader threads, one or more pre-compiler threads;
preparing data, via the one or more pre-compiler threads, by (i) fetching seed data, and (ii) storing the seed data in a user profile cache;
signaling, via the one or more pre-compiler threads, one or more signature matrix compiler threads;
building, based upon the seed data and the signaling and via the one or more signature matrix compiler threads, one or more signature matrices;
processing a first event of the stream of events based upon the one or more signature matrices, wherein the first event is associated with a first user; and
determining a degree of similarity between the first user and one or more members of the lookalike audience based upon the processing.

16. The non-transitory computer-readable storage medium of claim 15 , wherein the building comprises:
building a first signature matrix based upon two or more lookalike audiences.

17. The non-transitory computer-readable storage medium of claim 16 , the operations comprising:
determining a first result based upon a first lookup to the first signature matrix for the first event;
determining a second result based upon a second lookup to a second signature matrix; and
determining the lookalike audience based upon a combination of the first result and the second result.

18. The non-transitory computer-readable storage medium of claim 16 , the operations comprising:
updating the two or more lookalike audiences.

19. The non-transitory computer-readable storage medium of claim 18 , the operations comprising:
building a second signature matri

Assignees

• Yahoo Holdings Inc
• Oath Inc

Inventors

• Luo Haiyan
• Chen Datong
• Xia Zhen
• Cheng Shiyong
• Mao Yi
• Zhang Shu
• Wen Jiayi
• Cheng Xiaojiang
• Marcellini Herve

Classifications

• H04L67/306

  User profiles · CPC title

• G06Q30/0277

  Online advertisement · CPC title

• G06N20/20

  Ensemble learning · CPC title

• G06Q30/0269 Primary

  based on user profile or attribute · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title