Granted Patent B2

US 12,190,356 · App. 17/863,019 · Granted Jan 7, 2025

Inventors: Moussa Taifi (Jackson Heights, NY); Yana Volkovich (New York, NY); Carlos Eduardo Rodriguez Castillo (Brooklyn, NY)

Assignee: MICROSOFT TECHNOLOGY LICENSING, LLC

G06Q30/0277G06F18/23G06N20/00G06Q30/0256G06Q30/0261G06Q30/0275G06V10/764

Abstract

Claims (100)

1. A device, comprising:

a processing system including a processor; and

a memory that stores executable instructions that, when executed by the processing system, facilitate performance of operations, the operations comprising:

receiving, from a campaign manager device, information defining a line item and constraints for the line item in an online advertising system;

collecting browsing history information for targetable users;

ranking the targetable users based upon the browsing history information;

building a new segment based on the ranking of the targetable users;

forming a list of line item, segment identifier pairs, where the list includes a pair that comprises an identifier for the line item and an identifier for the new segment;

constructing a mapping that maps line items to segments, where the mapping is constructed based upon the list of line item, segment identifier pairs;

generating a line item to segment co-occurrence matrix based upon the mapping;

training a machine learning model to output an embedding of a segment identifier to add to the line item, where the machine-learning model is trained based upon the line item to segment co-occurrence matrix;

receiving, from the machine learning model, the embedding of the segment identifier;

receiving a second embedding of a second segment identifier that identifies a second segment;

selecting the second segment based upon a proximity between the embedding of the segment identifier output by the machine learning model and the second embedding of the second segment identifier;

providing advertisement content to targeted users according to the line item including the second segment.

2. The device of claim 1 , the operations further comprising:

producing a segment embeddings space responsive to training the machine learning model.

3. The device of claim 2 , the operations further comprising generating a second machine learning model to rank the targetable users, where generating the second machine learning model comprises:

computing co-occurrence for the targetable users based on universal resource locator (URL) information to generate user-user co-occurrence pairs;

training a user embedding model, including the segment embeddings space, with the user-user co-occurrence pairs, forming a trained user embedding model; and

storing the trained user embedding model.

4. The device of claim 3 , wherein the training the user embedding model comprises training a machine learning model that encodes user browser histories into the segment embeddings space where similarly behaving users cluster together.

5. The device of claim 4 , the operations further comprising:

using historical data to train the user embedding model for all targetable users that the line item would target if it did not have the constraints for the line item, forming a trained user embedding model.

6. The device of claim 5 , the operations further comprising:

determining, from the constraints for the line item, a targeted audience for the line item;

projecting the targeted audience into the trained user embedding model;

computing an average of the targeted audience in the trained user embedding model;

ranking two or more targetable users based on distance of each targetable user from the average of the targeted audience in the trained user embedding model;

selecting a predetermined number of ranked targetable users; and

creating the new segment including the predetermined number of ranked targetable users.

7. The device of claim 2 , the operations further comprising:

receiving, from the campaign manager device, a list of segment identifiers;

providing the list of segment identifiers to the segment embeddings space;

receiving from the segment embeddings space an output list of segment identifiers having locations in the segment embeddings space closest to segments of the list of segment identifiers;

ranking the output list of segment identifiers; and

providing a second predetermined number of segment identifiers of the ranked output list of segment identifiers to the campaign manager device as a list of recommended segments to add to the line item.

8. The device of claim 1 , the operations further comprising:

receiving from the campaign manager device an indication to attach the new segment to the line item.

9. A method performed by a processing system that includes a processor, the method comprising:

receiving, from a campaign manager device that is in network communication with the processing system, information defining a line item and information comprising constraints for the line item in an online advertising system;

collecting browsing history information for targetable users;

ranking the targetable users based upon the browsing history information to form a ranked list of targetable users;

constructing a new segment based on the ranked list of targetable users;

forming a list of pairs, where a pair in the list of pairs includes an identifier for the line item and an identifier for the new segment;

constructing a mapping that maps line items line items to segments, where the mapping is constructed based upon the list of pairs;

constructing a line item to segment co-occurrence matrix based upon the mapping;

training a machine learning model based upon the line item to segment co-occurrence matrix, where the machine learning model is trained to output an embedding of a segment identifier that is to be added to the line item;

receiving, from the machine learning model, the embedding of the segment identifier in response to providing the machine learning model with the line item;

receiving a second embedding of a second segment identifier that identifies a second segment;

computing a proximity between the embedding of the segment identifier and the second embedding of the second segment identifier;

selecting the second segment based upon the computed proximity between the embedding of the segment identifier and the second embedding of the second segment identifier;

including the second segment in the line item in response to selecting the second segment; and

providing advertisement content to targeted users based upon the line item including the second segment.

10. The method of claim 9 , further comprising:

producing a segment embeddings space responsive to training the machine learning model.

11. The method of claim 10 , further comprising generating a second machine learning model to rank the targetable users, where generating the second machine learning model comprises:

computing co-occurrence for the targetable users based on universal resource locator (URL) information to generate user-user co-occurrence pairs;

training a user embedding model, including the segment embeddings space, with the user-user co-occurrence pairs, thereby forming a trained user embedding model; and

storing the trained user embedding model.

12. The method of claim 11 , where training the user embedding model comprises training a machine learning model that encodes user browser histories into the segment embeddings space where similarly behaving users cluster together.

13. The method of claim 12 , further comprising:

using historical data to train the user embedding model for all targetable users that the line item would target if it did not have the constraints for the line item, forming a trained user embedding model.

14. The method of claim 13 , further comprising:

determining, from the information including the constraints for the line item, a targeted audience for the line item;

projecting the targeted audience into the trained user embedding model;

computing an average of the targeted audience in the trained user embedding model;

ranking two or more targetable users based on distance of each targetable user from the average of the targeted audience in the trained user embedding model;

selecting a predetermined number of ranked targetable users; and

creating the new segment including the predetermined number of ranked targetable users.

15. The method of claim 10 , the further comprising:

receiving, from the campaign manager device, a list of segment identifiers;

providing the list of segment identifiers to the segment embeddings space;

receiving from the segment embeddings space an output list of segment identifiers having locations in the segment embeddings space closest to segments of the list of segment identifiers;

ranking the output list of segment identifiers; and

providing a second predetermined number of segment identifiers of the ranked output list of segment identifiers to the campaign manager device as a list of recommended segments to add to the line item.

16. The method of claim 9 , further comprising:

receiving from the campaign manager device an indication to attach the new segment to the line item.

17. A non-transitory, machine-readable medium, comprising executable instructions that, when executed by a processing system including a processor, facilitate performance of operations, the operations comprising:

receiving, from a campaign manager device that is in network communication with the processing system, information defining a line item and information comprising constraints for the line item in an online advertising system;

collecting browsing history information for targetable users;

ranking the targetable users based upon the browsing history information to form a ranked list of targetable users;

constructing a new segment based on the ranked list of targetable users;

forming a list of pairs, where a pair in the list of pairs includes an identifier for the line item and an identifier for the new segment;

constructing a mapping that maps line items line items to segments, where the mapping is constructed based upon the list of pairs;

constructing a line item to segment co-occurrence matrix based upon the mapping;

training a machine learning model based upon the line item to segment co-occurrence matrix, where the machine learning model is trained to output an embedding of a segment identifier that is to be added to the line item;

receiving, from the machine learning model, the embedding of the segment identifier in response to providing the machine learning model with the line item;

receiving a second embedding of a second segment identifier that identifies a second segment;

computing a proximity between the embedding of the segment identifier and the second embedding of the second segment identifier;

selecting the second segment based upon the computed proximity between the embedding of the segment identifier and the second embedding of the second segment identifier;

including the second segment in the line item in response to selecting the second segment; and

providing advertisement content to targeted users based upon the line item including the second segment.

18. The non-transitory, machine-readable medium of claim 17 , the operations further comprising:

producing a segment embeddings space responsive to training the machine learning model.

19. The non-transitory, machine-readable medium of claim 18 , the operations further comprising generating a second machine learning model to rank the targetable users, where generating the second machine learning model comprises:

computing co-occurrence for the targetable users based on universal resource locator (URL) information to generate user-user co-occurrence pairs;

training a user embedding model, including the segment embeddings space, with the user-user co-occurrence pairs, thereby forming a trained user embedding model; and

storing the trained user embedding model.

20. The non-transitory, machine-readable medium of claim 19 , where training the user embedding model comprises training a machine learning model that encodes user browser histories into the segment embeddings space where similarly behaving users cluster together.

Assignments (2)

Continuity (2)

References Cited (28)